A Method for Inter-cell Interference Coordination and Load Balancing in Dense Heterogeneous Networks Based on Graph Neural Networks

Abstract

This invention relates to the field of communication system technology, specifically to a method for interference coordination and load balancing between cells in dense heterogeneous networks based on graph neural networks. The method includes the following steps: treating user terminals as first-type nodes; treating base stations as second-type nodes; constructing heterogeneous edges based on user-base station relationships and inter-base station interference relationships; outputting a heterogeneous network graph that integrates service spatiotemporal features; inputting the heterogeneous network graph into a dual-channel graph neural network model, including an interference coordination channel and a load balancing channel; performing cross-base station orthogonal allocation on spectrum units with gain values ​​higher than a preset gain threshold; and migrating users in batches to the first available base station in the target base station sequence according to priority. This invention introduces spectrum coupling weights through spectrum-aware graph convolution to calculate frequency-level interference reduction potential, and generates migration paths based on an attention mechanism that integrates user graph embedding, bandwidth fluctuation ratio, and spatial information, achieving bidirectional mining of multi-source heterogeneous features.

Description

Technical Field

[0001] This invention relates to the field of communication system technology, and in particular to a method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks. Background Technology

[0002] With the continuous evolution of 5G and subsequent generations of mobile communication networks, dense heterogeneous networks are widely deployed as a core architecture to improve system capacity, coverage, and service quality. In this architecture, different types of base stations (such as macro base stations, micro base stations, and pico base stations) are deployed irregularly and densely in urban environments to support diverse service scenarios (such as eMBB, URLLC, and mMTC). However, this highly dense and heterogeneous deployment structure also brings significant challenges in interference and load management:

[0003] On the one hand, due to the frequent coverage overlap of multi-level base stations in physical space, strong interference is easily triggered during spectrum resource reuse. Especially in hotspot areas with high spectrum reuse rates, frequency conflicts will seriously weaken the spectrum efficiency of the system. Existing interference coordination methods mostly rely on static spectrum division or neighboring cell interference maps, but it is difficult to perceive the dynamic impact of actual service distribution on the interference structure and lacks the ability to identify the root causes of spectrum conflicts in a fine-grained manner.

[0004] On the other hand, service traffic exhibits highly uneven characteristics in both user space and time dimensions, leading to an imbalance in load distribution among base stations. In traditional solutions, user migration strategies are often based on single-factor heuristic rules such as signal strength and average load, failing to comprehensively consider user service characteristics, target base station resource constraints, and migration link costs. This results in insufficient accuracy in migration decisions, easily causing migration failures or load bounces, and affecting overall network stability.

[0005] In recent years, Graph Neural Networks (GNNs) have gradually gained attention in communication network optimization tasks as a structured relationship modeling tool. However, most existing graph learning-based resource scheduling methods model the network as a homogeneous graph structure, which cannot simultaneously characterize the service coupling relationship between users and base stations as well as the interference relationship between base stations, thus limiting the model's ability to express multi-source dynamic information. Summary of the Invention

[0006] This invention provides a method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks. It integrates heterogeneous graph modeling methods that incorporate service, spatial, and spectrum information, and combines a dual-channel graph neural network structure to realize a systematic solution for collaborative optimization of spectrum interference coordination and load balancing, thereby improving the resource utilization efficiency and service quality assurance capabilities of dense heterogeneous networks.

[0007] A method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks includes the following steps:

[0008] S1. Construct a heterogeneous network graph with enhanced business features:

[0009] User terminals are designated as the first type of nodes; base stations are designated as the second type of nodes; heterogeneous edges are constructed based on user-base station association relationships and inter-base station interference relationships, wherein the heterogeneous edges include user-base station association edges and inter-base station interference edges;

[0010] Output a heterogeneous network diagram that integrates the spatiotemporal characteristics of integrated services;

[0011] S2. Dual-channel graph neural network joint decision-making: The heterogeneous network graph is input into a dual-channel graph neural network model, including an interference coordination channel and a load balancing channel, wherein:

[0012] Interference Coordination Channel: Through the spectrum sensing map convolutional layer, a spectrum efficiency gain distribution matrix for each base station is generated, which identifies the theoretical interference reduction magnitude of each spectrum unit;

[0013] Load balancing channel: Through the service migration graph attention layer, a load pressure transmission path is generated, and the migration priority and target base station sequence of users to be migrated are identified;

[0014] S3. Perform spectrum-load joint reconfiguration:

[0015] Based on the spectral efficiency gain distribution matrix, cross-base station orthogonal allocation is performed on spectral units with gain values ​​higher than a preset gain threshold;

[0016] Along the load pressure transmission path, users are migrated in batches to the first available base station in the target base station sequence according to priority.

[0017] Optionally, the node attributes of the first type of node include service type, location coordinates, and historical traffic fluctuation coefficient, wherein the historical traffic fluctuation coefficient is calculated based on the standard deviation of traffic over multiple past time slots.

[0018] Optionally, the node attributes of the second type of node include base station height, remaining fronthaul bandwidth, and available spectrum units.

[0019] Optionally, the edge weight of the user-base station association edge is calculated based on the weighted coupling of user service type and link quality index. By using a preset service type priority mapping function, the QoS level corresponding to different service types is converted into numerical weights and combined with the real-time signal-to-noise ratio between the user and the home base station to obtain a link quality index oriented towards service performance.

[0020] The edge weight of the interference edge between base stations is calculated based on the product of spatial overlap and spectrum overlap. The overlap of the coverage area is quantified by calculating the proportion of shared service users between the two base stations to the total number of covered users. The degree of spectrum resource sharing is calculated by the intersection-exchange ratio of the available spectrum sets of the two base stations. The edge weight of the interference edge between base stations reflects the interference intensity caused by the coupling of spatial topology and spectrum resources.

[0021] The final output heterogeneous network graph includes a set of user nodes, a set of base station nodes, a set of associated edges from users to base stations, a set of interference edges between base stations, a user node attribute matrix, a base station node attribute matrix, and an edge weight matrix.

[0022] Optionally, the interference coordination channel is used to model the spectral interference relationship between base stations. It employs a spectrum-aware graph convolution mechanism to perform dual-source neighborhood feature aggregation on base station nodes, fusing information from neighboring user nodes and neighboring base station nodes. The aggregation weights for user nodes are generated based on a service-aware attention function, while the aggregation weights for base station nodes are calculated based on a weighted average of spectral overlap and interference edge weights. After feature aggregation, a predicted spectral efficiency gain value for each base station in each available spectrum unit is generated. , indicating base station In the spectrum unit The theoretical interference reduction margin is calculated by combining the base station hidden state, spectrum ownership identifier, and strongest interference link information, and outputting the predicted spectrum efficiency gain value through linear prediction and activation function; all predicted spectrum efficiency gain values ​​constitute the spectrum efficiency gain distribution matrix.

[0023] Optionally, the load balancing channel models the feasibility and priority of user migration. Based on the graph attention mechanism, a migration utility scoring function is constructed between each user and the candidate target base station to obtain a migration utility score. The migration utility score integrates the graph embedding state of the user and the base station, the remaining fronthaul bandwidth of the target base station, the user traffic fluctuation coefficient, and the spatial distance. Based on the load rate and service type priority of the base station currently connected to the user, the migration priority is calculated, and the set of users to be migrated is selected.

[0024] Optionally, the load balancing channel further includes generating an ordered sequence of target base stations for each user to be migrated based on migration utility score and hop count decay mechanism, constructing a load pressure transmission path, and outputting a path set including user identifier, migration priority and target base station sequence for load balancing scheduling.

[0025] Optionally, S3 includes spectrum unit screening, specifically including calculating the global average value of the spectrum efficiency gain matrix and combining it with an adjustable coefficient to obtain a preset gain threshold under the current time slot, screening the spectrum efficiency gain value of each base station, retaining only spectrum units that are greater than the preset gain threshold, forming a set of high-value spectrum units of the current base station as candidate resources for priority scheduling.

[0026] Optionally, the cross-base station orthogonal allocation includes using all base stations as nodes. If two base stations have overlapping frequency points in the high-gain spectrum set, a connection edge is established between them to form a spectrum conflict graph. A graph coloring operation is performed on the conflict graph to group all base stations. It is required that there is no spectrum conflict between base stations in each group. The high-gain spectrum units in each group are uniformly assigned to the same orthogonal resource pool to ensure that adjacent base stations do not transmit on the same frequency point, thereby achieving interference avoidance and efficient reuse of spectrum resources.

[0027] Optionally, S3 further includes:

[0028] Migration Batch Division: Sort all users to be migrated from high to low according to migration priority, and prioritize users with high service level or high load area. Calculate the number of users that each migration batch can accommodate based on the available fronthaul bandwidth of the current network and the average bandwidth requirement of a single user. Divide all sorted users into several batches and perform the migration step by step in batches.

[0029] Target base station selection: For each batch of users, traverse the pre-generated target base station sequence and check the resource status of each candidate base station in turn. When the remaining fronthaul bandwidth of a base station exceeds the set safety threshold of the user's bandwidth requirement, and the number of its available spectrum units is not less than twice the user's requirement, the current base station is considered to have sufficient carrying capacity and is immediately marked as the user's migration target.

[0030] Migration operation: Allocate corresponding spectrum resources for the user in the orthogonal spectrum pool of the target base station, and complete the user context transfer and data plane switching through the interface.

[0031] The beneficial effects of this invention are:

[0032] This invention constructs a heterogeneous network graph that integrates user type, location, traffic fluctuations, base station bandwidth, and spectrum status. By introducing user-base station service weights and inter-base station interference edge weights, it achieves structured modeling of the relationship between service requirements, channel status, and spectrum resources in the network. Compared with traditional graph construction methods based solely on geographical distance or RSRP, the proposed graph structure significantly improves the graph neural network's ability to express interference propagation paths and load distribution patterns, providing a precise and dynamic network perception foundation for subsequent joint optimization.

[0033] This invention designs an interference coordination channel and a load balancing channel within a graph neural network structure. The former uses spectrum-aware graph convolution to introduce spectrum coupling weights to calculate frequency-level interference reduction potential, while the latter uses an attention mechanism to fuse user graph embeddings, bandwidth fluctuation ratios, and spatial information to generate migration paths, achieving bidirectional mining of multi-source heterogeneous features. Although the two channels model different tasks, they share the underlying graph structure and embedding space, ultimately forming an explicit linkage in spectrum allocation and user migration strategies, overcoming the limitation of traditional methods that model spectrum optimization and load migration separately.

[0034] In the execution phase, this invention accurately identifies and isolates conflicting frequency points through a gain threshold-driven spectrum filtering and graph coloring orthogonal allocation method, reducing inter-cell co-channel interference. At the same time, during user migration, a target base station selection strategy based on dual detection of fronthaul bandwidth and spectrum unit is introduced, and the instantaneous load of the system is controlled by dynamic migration batch division, effectively preventing chain overload and signaling storms caused by migration. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of dual-channel joint decision-making in an embodiment of the present invention. Detailed Implementation

[0038] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. For some well-known technologies, those skilled in the art may also use other alternative methods to implement the invention. Moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0039] like Figures 1-2 As shown, the method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks includes the following steps:

[0040] S1. Construct a heterogeneous network graph with enhanced business features:

[0041] User terminals are designated as the first type of nodes; base stations are designated as the second type of nodes; heterogeneous edges are constructed based on user-base station association relationships and inter-base station interference relationships, including user-base station association edges and inter-base station interference edges; and a heterogeneous network graph with fused service spatiotemporal characteristics is output.

[0042] S1 includes user node definitions: Active user terminals are considered as first-class nodes, and their node attributes include:

[0043] ① Service type: The service type of each user terminal is selected from the set {eMBB, URLLC, mMTC} and represented by a standardized identifier.

[0044] ② Position coordinates: The user's real-time position in a three-dimensional Cartesian coordinate system The location update period is no more than 100ms, used to support spatial interference mapping and topology change detection.

[0045] ③ Historical flow fluctuation coefficient: A fluctuation factor calculated in the form of standard deviation based on flow data from the past N time slots. ;in, Indicates the current time slot Historical flow fluctuation coefficient, The length of the sliding time window. For historical time slot indexes within a time window, This is the current timeslot number. For the first User traffic value per time slot Indicates the past The average flow rate per time slot is defined as: .

[0046] S1 includes the definition of base station nodes: macro base stations, micro base stations, and pico base stations within the target area are uniformly regarded as second-type nodes, and their node attributes include:

[0047] ① Base station height: The vertical height of the base station antenna phase center relative to the sea level, denoted as Used for fine modeling of 3D path loss. Indicates base station The vertical height of the antenna phase center relative to the sea level.

[0048] ② Remaining fronthaul bandwidth: The current unused fronthaul link transmission capacity, measured in Gbps, denoted as... , indicating base station The current remaining fronthaul link bandwidth.

[0049] ③ Available spectrum unit set: The set of physical resource blocks that the base station can dynamically schedule, defined as:

[0050] ;in, Indicates base station The currently available set of spectrum units, Indicates base station The One available spectrum unit, Indicates base station The number of currently available spectrum units (i.e., the set) (the base number).

[0051] The construction of heterogeneous edges in S1 can be subdivided into the following two categories:

[0052] User-Base Station Association Edge (Directed Edge): A directed edge is established between user u and its home base station b, and its edge weight is defined as follows: ;in:

[0053] This represents the edge weight between user u and base station b. Indicates user Business type, This represents the real-time signal-to-noise ratio between user u and base station b. Indicates user The business type priority weight mapping function takes the values ​​{eMBB,URLLC,mMTC}:

[0054] eMBB → 0.8: Enhanced Mobile Broadband primarily serves high-bandwidth services such as high-definition video and AR / VR. It has high speed requirements but strong tolerance for latency, so it is given a medium priority weight of 0.8.

[0055] URLLC→1.0: Ultra-Reliable Low Latency Communication, such as telemedicine, autonomous driving and other mission-critical services, are extremely sensitive to latency and reliability, and are therefore given the highest priority weight of 1.0.

[0056] mMTC → 0.6: Massive Machine Type Communication, such as IoT sensor networks, has a small traffic volume and is not sensitive to latency, so it is given the lowest priority weight of 0.6.

[0057] ② Interference edge between base stations (undirected edge): Interference edge between any two base stations When the overlap of their coverage areas exceeds 10%, an undirected edge is established, and the edge weight is defined as follows: ;in, They represent The user set covered The degree of overlap is defined as:

[0058] ; Indicated by base station The user set covered Indicated by base station The user set covered Indicates being base station and The user sets (intersections) covered simultaneously Indicates being base station or The set of users covered (union), The radix operator for sets indicates the number of elements in the set;

[0059] The spectrum overlap index is defined as follows:

[0060] ; Indicates base station The set of available spectrum units, Indicates base station The set of available spectrum units, Indicates base station and The intersection of spectrum units (shared available spectrum resources), Indicates base station and The union of spectral units (all the spectral resources involved in both), The radix operator for sets indicates the number of elements in the set.

[0061] Based on the above node and edge information, the real-time output heterogeneous network graph structure is as follows:

[0062] ;in, This represents the overall structure of a heterogeneous network graph. This represents the set of user nodes (i.e., all active user terminals). This represents a set of base station nodes (including macro base stations, micro base stations, and pico base stations). This represents the set of edges connecting a user to a base station. This represents the set of interference edges between base stations. This represents a user node attribute matrix, where rows correspond to user nodes, and columns include business type, 3D coordinates, historical traffic fluctuation coefficients, etc. This represents a base station node attribute matrix, where rows correspond to base station nodes, and columns represent base station height, remaining fronthaul bandwidth, available spectrum set, etc. This represents the edge weight matrix, containing all... and The numerical representation of is used for graph neural network computation.

[0063] S2. Dual-channel graph neural network joint decision-making: The heterogeneous network graph is input into a dual-channel graph neural network model, including an interference coordination channel and a load balancing channel, wherein:

[0064] Interference Coordination Channel: Through the spectrum sensing map convolutional layer, a spectrum efficiency gain distribution matrix for each base station is generated, which identifies the theoretical interference reduction magnitude of each spectrum unit;

[0065] Load balancing channel: Through the service migration graph attention layer, a load pressure transmission path is generated, and the migration priority and target base station sequence of users to be migrated are identified.

[0066] The dual-channel graph neural network joint decision-making in S2 takes the heterogeneous network graph output by S1 as input, constructs a dual-channel graph neural network model for interference coordination and load balancing, extracts node states and executes joint resource reconstruction decisions through multi-layer graph convolution and attention mechanisms.

[0067] S21, Interference Coordination Channel Handling Procedure:

[0068] S211, Spectrum Sensing Map Convolutional Layer Operation: For each base station node Perform the following neighborhood aggregation operation:

[0069]

[0070] in, Indicates the first Layer time base station The node embedding representation, Indicates the first Layer time base station The node embedding representation, Indicates the first Layer-adjacent base stations The node embedding representation, Indicates the first Layered users The node embedding representation, Indicates the first Layer spectrum-aware convolution weight matrix, Indicates base station The set of connected user nodes Indicates connection with base station The set of adjacent base stations with interfering edges. This indicates a modified linear unit activation function. This represents the user-base station attention weight, i.e., the user node aggregation weight. This represents the edge weight between user u and base station b. Indicates candidate users Edge weights between the base station b and base station b This represents the spectral coupling coefficient between base stations, i.e., the aggregation weight of base station nodes. Indicates base station b and base station Interference edge weights between Indicates base station b and base station The size of the spectral intersection between them This represents the maximum value of the two numbers and is used to normalize the intersection weights.

[0071] S212, Generate the spectral efficiency gain distribution matrix: for each base station Each spectrum unit Perform the following mapping:

[0072] ;

[0073] in, Indicates base station In the spectrum unit The theoretical interference reduction range, This represents the spectral gain mapping weight matrix. This represents the Sigmoid activation function. This represents the base station embedding state output by the last layer of the graph neural network. This represents the path loss compensation factor. Indicates the spectrum attribution indicator variable, if the spectrum unit belong The value is 1 if the value is 1, otherwise it is 0. For base station b, the set of available spectrum units. For adjacent base stations The set of available spectrum units, Indicates base station and Interference edge weights between This indicates that for all units containing spectral elements neighboring base stations Perform the maximum value operation. γ is the path loss compensation factor, which is derived by back-calculation based on the statistical mean of path loss between base stations in the wireless propagation model. Its function is to balance the influence of interference side weights at different distance scales. It can be obtained by regression fitting of path loss between different base station pairs in historical communication data. In actual networks, the empirical value of γ is 0.3–0.7, and it is initially set to γ=0.5.

[0074] Final output spectral efficiency gain matrix: ,in ; Represents the real number field. It is the number of base station nodes. It is the total number of all available spectral units (i.e., spectral dimension). Representation matrix In the spectrum unit of the medium base station b The theoretical interference reduction magnitude is the specific element value in the matrix.

[0075] The aforementioned spectrum-aware graph convolutional layer operation performs feature updates for base station nodes using a graph neural network. The core of this process is as follows: First, the base station node retains information from its own historical features. Then, it aggregates features from user nodes connected to it, with the aggregation weights determined by the edge weights between the user and the base station (reflecting link quality and service priority). Simultaneously, it also aggregates information from other base station nodes with which it has interference relationships. The aggregation weights combine interference intensity and spectrum overlap to more realistically depict the propagation path of spectrum interference.

[0076] This invention employs a design that enables each base station to perceive both the service demand characteristics of its subordinate users and the potential interference from neighboring base stations when updating its state in the graph neural network. Compared to traditional methods that only consider the aggregation of physical locations or fixed topologies between base stations, this method can more dynamically reflect the actual situation of interference propagation and spectrum coupling, thereby improving the accuracy of interference identification and suppression.

[0077] The spectral efficiency gain distribution matrix generates the theoretical interference reduction potential for each base station on each available spectrum unit. Specifically, it takes the final state of the base station's graph neural network, information on the availability of the current spectrum, and the maximum interference intensity with neighboring base stations on that spectrum as input; through a linear mapping layer and activation function, it outputs a normalized spectral gain score; all spectral gain scores are combined into a matrix to guide subsequent spectrum allocation operations.

[0078] This invention employs a design that fuses three types of information—graph embedding, spectrum attribution, and interference intensity—to obtain a more granular estimate of spectrum resource efficiency. The output gain matrix can be used to select the spectrum units that are most worthy of priority allocation under interference conditions, thereby achieving fine-grained regulation of spectrum resources for interference reduction. Compared to schemes that allocate spectrum based solely on global average interference or static thresholds, this design can more fully reflect the current state of the network and achieve dynamic adaptive spectrum coordination capabilities.

[0079] S22, Load Balancing Channel Processing Flow:

[0080] S221, Business Migration Diagram Notes Layer Operations: For each user node Calculate its migration utility score to the candidate base station:

[0081] ;

[0082] in, Indicates user to base station Utility score of transfer This represents the migration score weight matrix. This represents the user-base station embedding state output by the last layer of the graph neural network. Indicates base station Current remaining fronthaul bandwidth, Indicates user Historical flow fluctuation coefficient, Represents the spatial coordinates of the user and the base station. It represents Euclidean spatial distance.

[0083] Calculate migration priority based on current base station load rate and user service type:

[0084] ;

[0085] in, A priority mapping function for user business types. The current load rate of the base station to which the user is connected. This is the set of high-load users to be processed.

[0086] The migration utility score calculates the "cost-effectiveness" rating for a user migrating to a candidate base station, integrating four key types of information:

[0087] The graph embedding state of users and base stations reflects the global behavioral features that both have learned in the graph neural network.

[0088] Remaining fronthaul bandwidth of the target base station: This measures whether the base station can handle the data demands of newly migrated users.

[0089] User traffic fluctuation coefficient: indicates whether user services are stable; the smaller the fluctuation, the easier it is to migrate smoothly.

[0090] Spatial distance between the user and the target base station: the closer the migration path, the better the channel conditions, and the lower the migration cost.

[0091] These features are concatenated and input into a scoring function, which outputs a numerical score to quantify whether the transfer is "worthwhile".

[0092] This invention enables the load balancing process to no longer rely solely on signal strength or static distance as the basis for judgment. It introduces user traffic stability and target base station bandwidth constraints to avoid service quality degradation after migration. It achieves multi-factor fusion and refined modeling for migration benefit assessment, which is closer to the actual network operation.

[0093] After completing the migration score for each user, a priority function is introduced to determine which users should be migrated first and where they should be migrated to. Based on the load rate of the current connected base station and the user's service type, the urgency of migration for each user is calculated. A higher priority means that the user should be diverted as soon as possible.

[0094] Path construction function: Selects the target combination with the highest score and fewest hops from all candidate target base stations to form an ordered target sequence, making the migration more in line with the requirements of resource saving and network topology efficiency.

[0095] Path attenuation term: By introducing a hop count attenuation factor, "long path" migration is suppressed to avoid excessively long migration chains that cause transmission delays and network oscillations.

[0096] This invention ensures that high-priority users in high-load base stations are processed first, improving the "pressure relief efficiency" of the migration strategy. Instead of selecting only the nearest or least idle base station, it combines capacity and dynamic evaluation of the overall network structure to realize a load pressure gradient release mechanism under multi-hop path planning. This is a key improvement that distinguishes it from the "one-time on-site migration" in existing solutions.

[0097] S222, Construct the load pressure transmission path: according to Select the set of users to be migrated in descending order. For each Generate target base station sequence ( (This is the index variable for migrating users).

[0098] ,in ;in, Represented as user The constructed target base station sequence, Indicates user The sequence of candidate migrating base stations, where M is the number of candidate base stations. Represents the set of all possible target base stations. Indicates the distance from the current base station to the candidate base station The number of jumps, This is the path decay factor, with a value range of [0.1, 1.0]. In most dense scenarios, it is recommended to initially set it to 0.3.

[0099] Final output path set:

[0100] ;in, This represents a triplet record (user, priority, target base station sequence) in the migration path set.

[0101] Spectrum gain mapping weight matrix The interference reduction estimate used to map base station hidden state characteristics and interference environment information to spectrum units is obtained as follows:

[0102] (a) Constructing a training dataset: Collect a large number of real or simulated network state snapshots, including the spectrum allocation status of each base station, graph embedding features, spectrum ownership information and neighboring base station interference indicators, and label the actual spectrum efficiency changes of each spectrum unit (as a supervision target).

[0103] (b) Initialize the model structure: Design the backbone of the graph neural network, and the output layer uses a linear transformation module to receive the concatenated vectors. and output the predicted value. .

[0104] (c) Using a regression loss function: The mean square error between the predicted value and the labeled true spectral efficiency improvement value is used as the loss function for backpropagation optimization.

[0105] (d) After multiple rounds of training, the converged linear mapping weights are... It can be used to generate spectrum allocation strategies during the inference phase.

[0106] Migration score weight matrix The method used to map the joint feature vector between the user and the target base station to a migration utility score is as follows:

[0107] (a) Constructing a migration sample dataset: Based on historical network load change records, label which user migration behaviors are successful and effective (i.e., improved load balance and QoS satisfaction after migration), and extract features such as graph embedding, bandwidth-fluctuation ratio, and spatial distance for each user-target base station pair.

[0108] (b) Constructing the scoring model: Concatenate the above features into a unified vector, use it as the input to the scoring function, and use a single-layer linear mapping structure as the output layer, the parameters of which are... .

[0109] (c) Training phase: Supervised learning is adopted, using the labels of whether the transfer is successful or not to perform binary classification cross-entropy loss, or reinforcement learning is adopted, using the degree of improvement of network load state as a reward signal to optimize the scoring function parameters.

[0110] (d) The weight matrix obtained after the final training convergence. Used to score the migration feasibility of any user-base station combination in online scenarios, and to assist in load balancing path planning.

[0111] S3. Perform spectrum-load joint reconfiguration:

[0112] Based on the spectral efficiency gain distribution matrix, cross-base station orthogonal allocation is performed on spectral units with gain values ​​higher than a preset gain threshold;

[0113] Along the load pressure transmission path, users are migrated in batches to the first available base station in the target base station sequence according to priority.

[0114] S3 performs spectrum-load joint reconstruction based on the joint output of the interference coordination channel and the load balancing channel. It performs cross-base station orthogonal reconstruction of spectrum units and user bandwidth migration control to achieve coordinated optimization of network interference suppression and resource load balancing.

[0115] S31, Orthogonal allocation of spectrum units:

[0116] S311, Gain Threshold Filtering: Sets the preset gain threshold for spectral efficiency gain. ;in, This is the spectral efficiency gain threshold. This represents an adjustable coefficient, reflecting the sensitivity to the current interference intensity. Gain distribution matrix The global mean of the gain distribution matrix is ​​used as a benchmark. This dynamic threshold design uses an adaptive reference line based on the overall spectral efficiency level to screen out spectrum units with true interference reduction potential, avoiding over-allocation or omission of high-quality resources due to a fixed threshold setting. Specifically, the global mean of the gain distribution matrix is ​​used as a benchmark to dynamically reflect the current interference environment and spectral efficiency level of the entire network; then, an adjustable coefficient is multiplied to enhance the flexibility of the screening, ensuring that the screened spectrum units have significantly higher-than-average scheduling value.

[0117] For each base station Screening for high-value spectral units: That is, all spectral efficiency gain values ​​in base station b are greater than the threshold. The set of spectrum elements, that is: from all spectrum elements of base station b, those frequency points with gain scores exceeding a set threshold are selected for subsequent priority allocation, where For the base station in the spectral efficiency gain matrix In the spectrum unit The predicted value.

[0118] S312, orthogonal allocation across base stations, treats each base station as a node in a graph. When two base stations have the same high-gain spectrum element, it indicates a potential interference conflict. Therefore, an edge is established between them, i.e.:

[0119] Constructing a spectrum collision graph: Among them, node set For all base stations, edge sets When any two base stations satisfy: When high-gain spectrum units conflict, an edge is established between them. A graph coloring algorithm is applied to the conflict graph, where each color represents an orthogonal spectrum resource group. Adjacent base stations cannot have the same color, and each color represents an independent spectrum resource group, which is equivalent to an "orthogonal channel set".

[0120] Finally, the base station groups corresponding to each color will be... Spectrum units are uniformly allocated to an orthogonal resource pool to ensure that spectrum reuse conflicts are avoided in spatially adjacent areas.

[0121] S32, User migration execution:

[0122] S321, Migration Batch Division: Divide the set of users to be migrated Sort by priority in descending order:

[0123] ;

[0124] satisfy: That is, the first The migration priority of each user should not be lower than that of the first user. Priority for each user.

[0125] After sorting, not all users are migrated at once. Instead, the number of users that can be migrated simultaneously, i.e., the capacity of a migration "batch," is calculated based on the available fronthaul bandwidth resources of the current network. Specifically, the total available fronthaul bandwidth is divided by the bandwidth required to migrate a single user to obtain the maximum allowed batch size. The entire user set is then divided into several migration batches according to this size. This division avoids a large number of users migrating simultaneously in a short period, which could cause signaling storms or instantaneous overload of the target base station, and adapts to the current system resource status; that is: setting the migration batch size: ;in, This indicates the total available fronthaul bandwidth on the network. This represents the average fronthaul bandwidth required for a single user migration. This indicates the floor function.

[0126] Divide the sorted user set into Each migration batch is executed sequentially. This indicates rounding up, meaning rounding up the total number of users. According to the maximum per batch When grouping people, the required number of batches, even if the last group has insufficient members. They should also be counted as a separate batch.

[0127] S322, Target base station selection, for the first Each user in each migration batch Along the target base station sequence:

[0128] ;

[0129] Sequential detection is used during user migration. A candidate target base station sequence is pre-generated for each user to be migrated, representing potential destinations. Each base station in this sequence is checked sequentially to determine if it has sufficient resources to accommodate the user. A base station is selected as the final migration target if it meets two conditions:

[0130] Sufficient fronthaul bandwidth: The remaining fronthaul bandwidth of the base station must be greater than or equal to a certain percentage threshold of the user's bandwidth requirements (usually used to ensure redundancy after migration and avoid immediate overload).

[0131] Sufficient spectrum resource redundancy: The number of available spectrum units for this base station is no less than twice the spectrum required by the user, in order to ensure the flexibility and stability of resource scheduling.

[0132] Once the first base station that meets both conditions is found, the system selects it as the user's migration target and stops traversing subsequent candidates.

[0133] Specifically, this means selecting the first base station that meets the following conditions. :

[0134] ;

[0135] in, Indicates base station Current remaining fronthaul bandwidth, Indicates user The bandwidth requirements of the business This represents the bandwidth margin threshold, with a value of 1.3, indicating that the remaining fronthaul bandwidth of the target base station should be at least 130% of the user's service bandwidth requirements. Indicates base station The current set of available spectrum units, Indicates user The required number of spectrum units. If the condition is met, the base station is marked as the user's migration target.

[0136] S323, Migration Operation:

[0137] Perform the following migration operations on the selected target base station:

[0138] To allocate the required spectrum units to migrating users in their orthogonal resource pool;

[0139] Switching between user context and data plane is accomplished via the X2 interface;

[0140] Update the load rate, remaining bandwidth, and spectrum usage of the source and target base stations.

[0141] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0142] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks, characterized in that, Includes the following steps: S1. Construct a heterogeneous network graph with enhanced business features: User terminals are designated as the first type of nodes; base stations are designated as the second type of nodes; heterogeneous edges are constructed based on user-base station association relationships and inter-base station interference relationships, wherein the heterogeneous edges include user-base station association edges and inter-base station interference edges; Output a heterogeneous network diagram that integrates the spatiotemporal characteristics of integrated services; S2. Dual-channel graph neural network joint decision-making: The heterogeneous network graph is input into a dual-channel graph neural network model, including an interference coordination channel and a load balancing channel, wherein: Interference Coordination Channel: Through the spectrum sensing map convolutional layer, a spectrum efficiency gain distribution matrix for each base station is generated, which identifies the theoretical interference reduction magnitude of each spectrum unit; Load balancing channel: Through the service migration graph attention layer, a load pressure transmission path is generated, and the migration priority and target base station sequence of users to be migrated are identified; S3. Perform spectrum-load joint reconfiguration: Based on the spectral efficiency gain distribution matrix, cross-base station orthogonal allocation is performed on spectral units with gain values ​​higher than a preset gain threshold; Along the load pressure transmission path, users are migrated in batches to the first available base station in the target base station sequence according to priority.

2. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 1, characterized in that, The node attributes of the first type of node include service type, location coordinates, and historical traffic fluctuation coefficient, which is calculated based on the standard deviation of traffic over multiple past time slots.

3. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 2, characterized in that, The node attributes of the second type of node include base station height, remaining fronthaul bandwidth, and available spectrum units.

4. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 3, characterized in that, The edge weight of the user-base station association edge is calculated based on the weighted coupling of user service type and link quality index. Through a preset service type priority mapping function, the QoS level corresponding to different service types is converted into numerical weights and combined with the real-time signal-to-noise ratio between the user and the home base station to obtain the link quality index oriented towards service performance. The edge weight of the interference edge between base stations is calculated based on the product of spatial overlap and spectrum overlap. The overlap of the coverage area is quantified by calculating the proportion of shared service users between the two base stations to the total number of covered users. The degree of spectrum resource sharing is calculated by the intersection-exchange ratio of the available spectrum sets of the two base stations. The edge weight of the interference edge between base stations reflects the interference intensity caused by the coupling of spatial topology and spectrum resources. The final output heterogeneous network graph includes a set of user nodes, a set of base station nodes, a set of associated edges from users to base stations, a set of interference edges between base stations, a user node attribute matrix, a base station node attribute matrix, and an edge weight matrix.

5. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 1, characterized in that, The interference coordination channel is used to model the spectral interference relationship between base stations. It employs a spectrum-aware graph convolution mechanism to perform dual-source neighborhood feature aggregation on base station nodes, fusing information from neighboring user nodes and neighboring base station nodes. User node aggregation weights are generated based on a service-aware attention function, while base station node aggregation weights are calculated based on a weighted average of spectral overlap and interference edge weights. After feature aggregation, a predicted spectral efficiency gain value for each base station in each available spectrum unit is generated. , indicating base station In the spectrum unit The theoretical interference reduction margin is calculated by combining the base station hidden state, spectrum ownership identifier, and strongest interference link information, and outputting the predicted spectrum efficiency gain value through linear prediction and activation function; all predicted spectrum efficiency gain values ​​constitute the spectrum efficiency gain distribution matrix.

6. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 1, characterized in that, The load balancing channel models the feasibility and priority of user migration. Based on the graph attention mechanism, a migration utility scoring function is constructed between each user and the candidate target base station to obtain a migration utility score. The migration utility score integrates the graph embedding state of the user and the base station, the remaining fronthaul bandwidth of the target base station, the user traffic fluctuation coefficient, and the spatial distance. Based on the load rate and service type priority of the base station currently connected to the user, the migration priority is calculated, and the set of users to be migrated is selected.

7. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 6, characterized in that, The load balancing channel also includes generating an ordered sequence of target base stations for each user to be migrated based on migration utility score and hop count decay mechanism, constructing a load pressure transmission path, and outputting a path set including user identifier, migration priority and target base station sequence for load balancing scheduling.

8. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 1, characterized in that, S3 includes spectrum unit screening, specifically including calculating the global average value of the spectrum efficiency gain matrix and combining it with an adjustable coefficient to obtain a preset gain threshold for the current time slot, screening the spectrum efficiency gain value of each base station, retaining only spectrum units that are greater than the preset gain threshold, forming a set of high-value spectrum units for the current base station as candidate resources for priority scheduling.

9. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 8, characterized in that, The cross-base station orthogonal allocation involves using all base stations as nodes. If two base stations have overlapping frequency points in the high-gain spectrum set, a connection edge is established between them to form a spectrum conflict graph. A graph coloring operation is performed on the conflict graph to group all base stations. It is required that there are no spectrum conflicts between base stations in each group. The high-gain spectrum units in each group are uniformly assigned to the same orthogonal resource pool to ensure that adjacent base stations do not transmit on the same frequency point, thereby achieving interference avoidance and efficient reuse of spectrum resources.

10. The method for inter-cell interference coordination and load balancing in dense heterogeneous networks based on graph neural networks according to claim 9, characterized in that, S3 further includes: Migration Batch Division: Sort all users to be migrated from high to low according to migration priority, and prioritize users with high service level or high load area. Calculate the number of users that each migration batch can accommodate based on the available fronthaul bandwidth of the current network and the average bandwidth requirement of a single user. Divide all sorted users into several batches and perform the migration step by step in batches. Target base station selection: For each batch of users, traverse the pre-generated target base station sequence and check the resource status of each candidate base station in turn. When the remaining fronthaul bandwidth of a base station exceeds the set safety threshold of the user's bandwidth requirement, and the number of its available spectrum units is not less than twice the user's requirement, the current base station is considered to have sufficient carrying capacity and is immediately marked as the user's migration target. Migration operation: Allocate corresponding spectrum resources for the user in the orthogonal spectrum pool of the target base station, and complete the user context transfer and data plane switching through the interface.

Images