Heterogeneous network intelligent slice resource scheduling method and device, electronic equipment and medium

Abstract

The invention relates to the technical field of network resource scheduling, and provides a heterogeneous network intelligent slice resource scheduling method and device, electronic equipment and a medium. The method comprises the following steps: performing three-dimensional joint modeling on heterogeneous network physical resources to obtain a dynamic resource state tensor, and performing a dynamic resource allocation decision by combining a deep reinforcement learning model with a service demand matrix to obtain an initial slice resource allocation matrix; and performing mixed integer linear programming through a preemption decision model in combination with the URLLC service request queue to obtain a resource allocation sequence, and performing occupancy rate updating on the dynamic resource state tensor according to the resource allocation sequence to obtain ground network resource data, and carrying out resource fusion in combination with the satellite beam capacity data to obtain a space-ground integrated resource pool tensor, thereby carrying out traffic prediction in combination with historical traffic data through a long-short memory network to obtain a bandwidth pre-adjustment strategy. Through organic integration of technologies such as mixed integer linear programming and traffic prediction, the overall network resource utilization rate is improved.

Description

Technical Field

[0001] This application relates to the technical field of network resource scheduling, and in particular, to a method, device, electronic device, and medium for intelligent slice resource scheduling in a heterogeneous network. Background Art

[0002] Meeting the growing and diverse service requirements under limited physical resources and ensuring the stability and efficiency of network operation have become important functions of network resource scheduling in modern communication networks.

[0003] Traditional heterogeneous network resource scheduling methods usually rely on preset rules or static allocation strategies, and often achieve resource scheduling through fixed time-frequency domain partitioning or simple priority sorting. There are certain limitations in the allocation of network resources by such methods. Their scheduling decision-making process lacks sufficient intelligence and flexibility, and cannot comprehensively integrate terrestrial and satellite resources, resulting in problems such as uneven resource allocation, response delay, and resource waste when the network state changes rapidly and service demands surge. Summary of the Invention

[0004] In view of this, this application provides a method, device, electronic device, and medium for intelligent slice resource scheduling in a heterogeneous network to solve the problem of being difficult to capture the dynamic changes of network resource status in real time and effectively respond to sudden service demands.

[0005] The first aspect of this application provides a method for intelligent slice resource scheduling in a heterogeneous network, and the method includes: Performing three-dimensional joint modeling processing on the preset physical resources of the heterogeneous network in the time domain, frequency domain, and space domain to obtain a dynamic resource status tensor; Performing dynamic resource allocation decision processing on the dynamic resource status tensor and the received service demand matrix through a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix; Performing mixed integer linear programming processing on the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence; Performing occupancy rate update processing on the dynamic resource status tensor according to the resource allocation sequence to obtain terrestrial network resource data, and performing resource fusion processing on the terrestrial network resource data and the received satellite beam capacity data to obtain a space-earth integrated resource pool tensor; Performing traffic prediction processing through a preset long short-term memory network according to the preset historical traffic data and the space-earth integrated resource pool tensor to obtain a bandwidth pre-adjustment strategy.

[0006] In an optional implementation, the three-dimensional joint modeling process of the preset heterogeneous network physical resources in the time domain, frequency domain, and spatial domain to obtain the dynamic resource state tensor includes: Performing time slot division processing on the heterogeneous network physical resources according to a preset time window to obtain a time slot set; Performing subcarrier division and priority weighting processing on the heterogeneous network physical resources to obtain a weighted frequency band set; Performing coverage area grid processing on the heterogeneous network physical resources according to the preset base station and satellite coverage areas to obtain a spatial unit set; Performing three-dimensional joint construction processing on the time slot set, the weighted frequency band set, and the spatial unit set to obtain the dynamic resource state tensor.

[0007] In an optional implementation, the dynamic resource allocation decision processing of the dynamic resource state tensor and the received service demand matrix through a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix includes: Performing joint encoding processing on the service demand matrix to obtain a state vector; Performing time-domain-frequency-domain-space three-dimensional resource dynamic allocation processing on the state vector and the dynamic resource state tensor through the deep reinforcement learning model to obtain the initial slice resource allocation matrix.

[0008] In an optional implementation, the hybrid integer linear programming processing of the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence includes: Performing urgency sorting processing on the URLLC service request queue to obtain a priority sequence; Performing hybrid integer linear function solution processing on the priority sequence through the preemption decision model to obtain a resource preemption flag vector and a path reconstruction instruction set; Updating the initial slice resource allocation matrix according to the resource preemption flag vector and the path reconstruction instruction set to obtain the resource allocation sequence.

[0009] In an optional implementation, the occupancy rate update processing of the dynamic resource state tensor according to the resource allocation sequence to obtain ground network resource data, and the resource fusion processing of the ground network resource data and the received satellite beam capacity data to obtain a space-earth integrated resource pool tensor includes: Performing resource occupancy rate extraction processing on the resource allocation sequence to obtain a resource occupancy rate parameter; Update the dynamic resource status tensor according to the resource occupancy rate parameter to obtain the terrestrial network resource data; Perform virtualization mapping processing on the satellite beam capacity data to obtain a set of virtual resource blocks; Merge the terrestrial network resource data with the set of virtual resource blocks to obtain the space-ground integrated resource pool tensor.

[0010] In an alternative embodiment, the traffic prediction processing by a preset long short-term memory network according to the preset historical traffic data and the space-ground integrated resource pool tensor to obtain a bandwidth pre-adjustment strategy includes: Normalize the historical traffic data to obtain a historical traffic input sequence; Perform traffic prediction processing on the historical traffic input sequence and the space-ground integrated resource pool tensor through the long short-term memory network to obtain traffic prediction data; Perform sliced bandwidth dynamic adjustment processing on the traffic prediction data according to the average traffic in the historical traffic data to obtain the bandwidth pre-adjustment strategy.

[0011] In an alternative embodiment, the method further includes: Perform local resource gap calculation processing according to the bandwidth pre-adjustment strategy and the space-ground integrated resource pool tensor to obtain a resource gap amount; When the resource gap amount is greater than zero, construct transaction request data for the resource gap amount according to a preset transaction priority coefficient to obtain a cross-operator resource transaction request; Perform operator offer matching processing on the cross-operator resource transaction request according to a preset blockchain smart contract to obtain a resource allocation encryption certificate; Update the space-ground integrated resource pool tensor according to the resource allocation encryption certificate and the resource gap amount.

[0012] A second aspect of the present application provides a heterogeneous network intelligent slice resource scheduling device, the device includes: A three-dimensional joint module, configured to perform three-dimensional joint modeling processing on preset heterogeneous network physical resources in the time domain, frequency domain, and spatial domain to obtain a dynamic resource status tensor; A resource allocation module, configured to perform dynamic resource allocation decision processing on the dynamic resource status tensor and the received service demand matrix through a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix; A linear programming module for performing mixed-integer linear programming processing on the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence; A resource fusion module for performing occupancy rate update processing on the dynamic resource status tensor according to the resource allocation sequence to obtain ground network resource data, and performing resource fusion processing on the ground network resource data and the received satellite beam capacity data to obtain a space-ground integrated resource pool tensor; A traffic prediction module for performing traffic prediction processing through a preset long short-term memory network according to preset historical traffic data and the space-ground integrated resource pool tensor to obtain a bandwidth pre-adjustment strategy.

[0013] The third aspect of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the heterogeneous network intelligent slice resource scheduling method described above are implemented.

[0014] The fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the heterogeneous network intelligent slice resource scheduling method described above are implemented.

[0015] In summary, the present application at least includes the following beneficial technical effects: 1. By making full use of the advantages of combining deep learning and mathematical programming, it can quickly respond to the dynamically changing network environment, especially showing a high response speed and resource allocation flexibility when facing the scheduling requirements of emergency services.

[0016] 2. Through the construction of the space-ground integrated resource pool tensor and refined traffic prediction, various resources can be allocated more reasonably, resource waste can be reduced, and at the same time, the timely allocation of key resources can be ensured when emergency services occur. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of a heterogeneous network intelligent slice resource scheduling method provided by an embodiment of the present application; Figure 2It is a functional module diagram of a heterogeneous network intelligent slice resource scheduling device provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0020] As Figure 1 shown, it is a flowchart of a heterogeneous network intelligent slice resource scheduling method provided by an embodiment of the present application. The heterogeneous network intelligent slice resource scheduling method provided by an embodiment of the present application includes the following steps.

[0021] Step S1: Perform three-dimensional joint modeling processing on the preset heterogeneous network physical resources in the time domain, frequency domain, and spatial domain to obtain a dynamic resource state tensor.

[0022] It should be understood that the network resource state is periodic and bursty over time. Time slot division can quantitatively describe the distribution of resources in the time domain, which helps to precisely control dynamic resource scheduling and allocation. In the embodiments of the present application, all the physical resources in the heterogeneous network (i.e., heterogeneous network physical resources) are divided according to a preset time window, thereby forming a clear time period sequence. The divided time period is called a "time slot", and each time slot represents the available situation of network resources during that time period. Specifically, first, the real-time status data of the heterogeneous network physical resources is obtained through a network management platform or a monitoring system. The real-time status data includes the operating parameters of 5G base stations, the signaling information of satellite communication systems, and the status information of Internet of Things terminals. At the same time, the operator determines the time window in advance according to service requirements and network characteristics. For example, it can be set to 1 millisecond, 10 milliseconds, or a longer time period. The selection of the time window must consider the network service quality requirements and the characteristics of resource dynamic changes. After dividing the preset time window, each time slot obtained has a fixed time length and predefined start and end times in network resource scheduling. Subsequently, the entire time period is evenly divided according to the preset window. For example, the total time period is set as T and divided into n time slots to form a time slot set T = {t 1 , ……, t n}. Among them, each time slot t i represents that the network resources are in a relatively stable state during the corresponding time period. The resource capacity corresponding to each time slot is denoted as C t, which is used to reflect the available amount of resources during this period. For example, if the transmission resources, computing power, and spectrum resources of a base station reach a certain percentage within a time slot, then the C t value of this time slot can be calculated comprehensively from these data. It should be understood that the implementation of time slot division relies on precise clock synchronization technology (for example, Precision Time Protocol) to ensure time consistency among devices. After division, the formed set of time slots provides discretized information in the time domain, which helps to predict and optimize based on the resource conditions within each time slot. In an optional implementation manner, time slot division also takes into account the peaks and valleys of network traffic, associates the divided time slots with the network load conditions, and ensures that the statistical information of resources within each time slot accurately reflects the current network state.

[0023] Since different services have different requirements for spectrum resources, the weighted division of the spectrum part in the heterogeneous network physical resources in the frequency domain in the embodiments of the present application can make resource scheduling more in line with the actual service requirements and improve the effectiveness of the scheduling strategy. For example, due to the strict requirements for latency and reliability of the URLLC service, a higher weight can be given during frequency band division, while the eMBB service may obtain a relatively lower weight. Specifically, first, spectrum information is extracted from the heterogeneous network physical resources, and this information usually includes parameters such as frequency band range, frequency resolution, and actual available bandwidth. According to the preset spectrum planning scheme, the entire frequency band is divided into several subcarriers. The division method can be uniform division or non-uniform division based on actual needs, and the specific method depends on the spectrum utilization situation and service requirements of the network. After division, a set of subcarriers F={F 1 , ……, F m} is formed, and each subcarrier F i represents an independent and smaller frequency band unit in the spectrum. During the division process, an initial weight coefficient w f is assigned to each subcarrier, and the initial weight coefficient reflects the importance of this frequency band for a specific service. The key to priority weighting processing lies in assigning different weight coefficients to different frequency bands according to the preset service requirements (for example, URLLC, eMBB, mMTC). For example, to ensure that the URLLC service can obtain sufficient resources, the weight of the frequency band related to this service can be set to 1, while the weights of other services are set to values less than 1. Among them, the initial weight coefficient setting can be determined based on historical service data, real-time service requirements, and network resource utilization rate. The weighting process takes into account both the physical characteristics of the spectrum and the service quality requirements.

[0024] Meanwhile, in the actual network, there are significant differences in user density, traffic demand, and resource availability in different regions. Discretizing the spatial distribution information in the heterogeneous network's physical resources and dividing the geographical area into multiple grid cells can form a clear spatial distribution model, which provides a spatial reference basis for resource scheduling. Discretizing the region into multiple spatial units allows for individual assessment of the resource status in each region and enables refined management during the scheduling process. Specifically, first obtain the coverage ranges of base stations and satellites through geographic information systems and network coverage planning data. For base stations, the coverage range data usually comes from base station planning maps and signal strength distribution maps; for satellites, it is obtained based on on-board data and coverage information provided by ground stations. The preset coverage area is usually determined by geographical boundaries, administrative divisions, or signal strength thresholds. Based on this information, divide the entire service area into multiple smaller grid cells, with each cell representing a specific geographical region. Next, define each grid cell in the division process as a spatial unit to obtain a set of spatial units S = {S 1 , ……, S k}. Within each spatial unit, calculate the resource demand density ρ S to reflect the resource demand intensity in that region. The calculation of resource demand density usually depends on factors such as the number of users, traffic volume, and geographical area. When calculating the user density and service data in each grid cell using GIS data and real-time user statistics information, the boundaries of each spatial unit can also be standardized to ensure that the areas of all units are unified or distributed according to a predetermined ratio.

[0025] Finally, due to the coupling effect of the distribution of network resources in the time domain, frequency domain, and spatial domain, the three sets of discrete data obtained are jointly constructed to form a three-dimensional tensor (i.e., the dynamic resource state tensor) that describes the resource state of the heterogeneous network, which can more comprehensively capture the dynamic changes in resource distribution. Among them, each element of the dynamic resource state tensor reflects the available state of resources at a specific time, specific frequency band, and specific spatial unit, providing a global view for subsequent scheduling algorithms. Specifically, through the Cartesian product operation on the set of time slots, the set of weighted frequency bands, and the set of spatial units, that is, in each spatial unit within each frequency band within each time slot, there corresponds a resource state value. Among them, the resource state value is denoted as r t,f,s representing the remaining available amount of resources in time slot t, frequency band f, and spatial unit s. The constructed dynamic resource state tensor is denoted as .

[0026] During the data processing process, the r t,f,s value of each combination needs to be calculated based on actual measurement data and preset parameters. First, obtain the C of each time slot from the time slot division step tValue; secondly, obtain the weight w of each subcarrier from the weighted frequency band set f ; furthermore, obtain the demand density ρ of each grid from the spatial grid processing S ; finally, calculate the occupancy rate δ of each location according to the current resource allocation situation i(t,f,s) , so as to further calculate the remaining allocable resource ratio according to the occupancy rate. Finally, multiply the C t value, weight w f , demand density ρ S and the obtained allocable resource ratio to obtain the r value of each combination t,f,s , so as to construct a complete dynamic resource state tensor R. Each element of the dynamic resource state tensor R reflects the current available state of network resources in a specific time slot, frequency band and spatial unit, which can not only be used for real-time resource scheduling, but also be stored as historical data for subsequent prediction model training and scheduling strategy optimization.

[0027] Step S2: Perform dynamic resource allocation decision processing on the dynamic resource state tensor and the received service demand matrix through a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix.

[0028] Since the received service demand matrix itself is a multi-dimensional data structure, in order to fuse it with the dynamic resource state tensor, it must first be subjected to feature extraction and encoding to form a low-dimensional and descriptive state vector, so as to facilitate the subsequent model to consider the specific requirements of each service slice while jointly considering time-domain, frequency-domain and spatial-domain resource information. It should be understood that each row in the obtained service demand matrix corresponds to a network slice, and each slice contains three-dimensional demand data, namely bandwidth demand B i,req , delay demand L i,req and reliability demand R i,req . These data are obtained through network operator configuration or application layer negotiation, and are transmitted in a structured format (such as JSON or XML) during the transmission process.

[0029] First, before the demand data in the business demand matrix enters the encoding processing module, it will undergo preprocessing operations such as normalization and denoising to ensure consistent data ranges and no outliers. Next, the preprocessed business demand matrix is jointly encoded with other auxiliary information. Here, the auxiliary information includes various indicators in the business matrix and global features that may be obtained through statistical analysis, such as the average bandwidth demand of all slices, statistical quantities of the delay distribution, and reliability distribution. The process of joint encoding can adopt methods such as feature concatenation, linear transformation, and non-linear mapping. For example, each slice demand data vector in the original matrix is concatenated with the global statistical vector to generate an extended vector. After that, the encoding status vectors of all slices are integrated to form an overall status vector S, which is used to represent the distribution of business demands at the entire network slice level. The status vector S can be obtained by stacking the status vectors of all slices, i.e., S = [s 1 ; s 2 ; ……; s N . Among them, ";" represents the vertical concatenation of vectors, generating a matrix with a dimension of Nxd, where d is the dimension of each slice status vector after joint encoding. The status vector S, as one of the inputs of the deep reinforcement learning model, can simultaneously reflect the individual demands and overall trends of each slice, providing data support for the model when making resource allocation decisions.

[0030] During the encoding process, activation functions (such as ReLU or sigmoid) are used to perform non-linear mapping on the concatenated data to improve the expression ability of the status vector, enabling each component in the vector to better represent the non-linear characteristics of slice service demands. The encoding module can be designed as a feed-forward neural network. Its input layer receives the original concatenated vector and outputs the final status vector through a series of fully connected layers and non-linear activations. The weight parameters in this process can be pre-trained offline using historical data and supervision signals to ensure that the encoding results have high robustness and discriminative ability in different scenarios. In addition, during the joint encoding process, attention also needs to be paid to data scale matching and dimension conversion. Since there may be differences in the dimensions of the original demand data and statistical data, normalization techniques (such as min-max normalization or z-score standardization) are usually used to convert the data into a unified scale to prevent a certain piece of data from dominating the encoding due to its large value. The normalized data is input into the encoding network and then undergoes non-linear mapping to form the status vector. This can ensure that each input feature contributes evenly during joint encoding, facilitating subsequent dynamic resource allocation.

[0031] Finally, the encoded results are checked through data verification and visualization means to ensure that the state vector fully captures the key information in the business requirements and its distribution characteristics can reflect the differences between different slices. The encoded state vector not only transforms the original business requirements into a format that can be directly processed by the deep reinforcement learning algorithm but also enhances the robustness of the data to a certain extent, providing a solid foundation for subsequent dynamic resource scheduling based on the state vector.

[0032] Furthermore, network resources vary dynamically in the three dimensions of time, frequency, and space, while the business requirement state vector provides the demand information for each slice. Using the deep reinforcement learning model, an optimal resource allocation strategy can be learned to achieve the optimization of global resource scheduling. The deep reinforcement learning model jointly processes the state vector obtained from joint encoding and the previously constructed dynamic resource state tensor to generate an initial slice resource allocation matrix. The initial slice resource allocation matrix describes the resource ratio that each slice should obtain in each time slot, each frequency band, and each spatial unit. It should be understood that the dynamic resource state tensor R describes the real-time distribution of network resources within the time slot set T, the weighted frequency band set F, and the spatial unit set S; the state vector S ∈ R N×d reflects the business requirements of each slice. To enable the deep reinforcement learning model to consider both types of information simultaneously, it is first necessary to appropriately fuse these two parts of data. Usually, the state vector is extended to match the dimension of the dynamic resource state tensor, or the attention mechanism is used to organically combine the information of the two. Specifically, the deep reinforcement learning model adopts a joint feature fusion method, mapping the state vector through a fully connected neural network layer to the same feature space as the dynamic resource state tensor, and obtaining a new state vector S' after mapping.

[0033] Furthermore, the deep reinforcement learning model (such as DQN, DDPG, or Actor - Critic structure) receives the mapped state vector S' and the dynamic resource state tensor R as inputs. The model is internally designed with a specific network structure to extract local features in the time domain, frequency domain, and spatial domain through convolutional layers or three - dimensional convolutional layers, and at the same time capture the business requirement features in the state vector through fully connected layers. Finally, an initial slice resource allocation matrix is generated through the deep reinforcement learning model. Among them, each element a in the initial slice resource allocation matrix t,f,s represents the resource allocation ratio obtained in the time slot t, frequency band f, and spatial unit s. The initial slice resource allocation matrix reflects the preliminary decision of the deep reinforcement learning model on global resource scheduling.

[0034] It should be understood that during the training process of the deep reinforcement learning model, a reward function is used as the optimization objective. The design of the reward function needs to consider both the satisfaction of the service requirements of the slices and the resource utilization rate. The reward function adopted in the embodiments of this application is as follows: where R is the reward value, representing the overall performance of the deep reinforcement learning model under the current resource allocation. B i , L i , R i respectively represent the bandwidth, latency, and reliability metrics actually allocated to slice i. B i,req , L i,req , R i,req respectively represent the service requirement values corresponding to slice i. α, β, and γ are weight coefficients, and satisfy α + β + γ = 1. λ is the coefficient of the resource fragmentation penalty term.

[0035] In the decision-making process of deep reinforcement learning, the model continuously interacts with the environment, evaluates the rewards, and adjusts the internal parameters using the backpropagation algorithm to make the predicted initial slice resource allocation matrix achieve the optimal allocation effect. During the decision-making process, the model needs to effectively fuse the information in the state vector and the dynamic resource state tensor, so as to consider the individual needs of each slice and also reflect the distribution of resources in the time, frequency, and space dimensions. After multiple trainings and optimizations, the policy learned by the model will be able to automatically output the resource proportion to be allocated for each slice in each time slot, each frequency band, and each spatial unit.

[0036] Step S3: Perform mixed-integer linear programming processing on the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence.

[0037] Since there are significant differences in the urgency of each request in the network, in order to prevent service interruption of critical tasks due to insufficient resource allocation, it is necessary to clearly sort all requests so that high-urgency requests can be preferentially satisfied during preemption decision-making. In the embodiments of this application, the relative urgency of requests is sorted by quantifying the service level and latency deadline of each request. The sorting result will directly affect the subsequent resource preemption decision, ensuring that urgent services are preferentially scheduled when resources are limited, so as to meet the requirements of ultra-low latency and high reliability for Ultra Reliable Low Latency Communication (URLLC). Specifically, obtain the URLLC request queue Q URLLC ={q 1 ,……,q m} from the edge node or real-time monitoring system. Each request q jContains at least two key attributes: Service Level SL j and latency deadline Deadline j . The service level is usually represented by a numerical value, and the higher the value, the higher the priority; the latency deadline indicates how long the request needs to complete the response. To comprehensively consider these two factors at the same time, the embodiments of the present application use the following priority calculation formula to score each request: where p j represents the comprehensive priority score of the j-th request. SL j represents the service level of the j-th request, and a higher value indicates greater importance of the request. Deadline j represents the latency deadline of the j-th request, and a smaller value indicates that the request needs to be responded to faster, so the reciprocal value is larger. μ and ν are weight coefficients used to balance the importance of the service level and the deadline. Both of these coefficients are positive numbers and are preset according to actual needs.

[0038] After obtaining the priority score of each request, the queue Q URLLC is sorted in descending order of the score. The generated priority sequence P URLLC ={p 1 , ……, p m} determines which requests should be given priority when preempting resources. The sorting process uses standard sorting algorithms (such as quicksort or mergesort) to ensure the accuracy and efficiency of the sorting result. The sorted priority sequence provides a clear goal for the subsequent mixed-integer linear programming solution, that is, to maximize the total preemption benefit.

[0039] Furthermore, the preemption decision model set in the system constructs a mixed-integer linear programming mathematical model based on the obtained priority sequence, and obtains a resource preemption flag vector and the corresponding path reconstruction instruction by solving this model. That is, it is determined which requests should be preempted under the given resource constraints, and how to reallocate the resources after preemption to achieve the priority response of emergency services. The mixed-integer linear programming solution uses mathematical optimization methods to achieve global optimal or suboptimal decisions under limited resources. Specifically, assuming that the initial slice resource allocation matrix is , for each URLLC request, a mixed-integer linear programming mathematical model is constructed according to its slice and priority score. Among them, the decision variable x jIndicates whether the j-th request successfully preempts, with values of 0 or 1. The objective function of the constructed mixed-integer linear programming mathematical model is designed to maximize the total priority benefit of all emergency requests. By doing so, it ensures that under limited resource conditions, requests with high priorities are selected for resource preemption, thereby optimizing the overall service quality. At the same time, the mixed-integer linear programming mathematical model needs to introduce constraint conditions to ensure that the resources preempted within each time slot-frequency band-space unit do not exceed the resource status value capacity at that location (i.e., satisfy ). By solving the mixed-integer linear programming mathematical model, a set of binary decision variables X = [x 1 , ……, x m can be obtained. This vector is the resource preemption flag vector, indicating whether each request is selected for preemption. At the same time, the optimal solution generated during the solving process also provides a path reconstruction instruction set corresponding to each request. This instruction set defines how to adjust the transmission path in real time after resource preemption, such as updating the SDN flow table, modifying the satellite beam pointing, etc., to ensure that the service flow after preemption can be smoothly migrated to the newly configured path.

[0040] The preemption decision model uses existing mathematical optimization tools (such as CPLEX or Gurobi) to solve the mixed-integer linear programming mathematical model. Its solving process strictly follows the objective function and constraint conditions to ensure a globally optimal or near-optimal solution. After solving, the obtained resource preemption flag vector and path reconstruction instruction set are both saved as system parameters and passed to subsequent modules for further processing. Through mixed-integer linear programming, binary decision problems and their combinatorial constraints can be precisely processed to ensure maximizing the total priority of emergency requests under limited resources, thus making the network scheduling decision more targeted and reasonable. In actual data processing, when solving the mixed-integer linear programming mathematical model, the dynamic changes of each resource unit need to be considered, and the resource status value r t,f,s needs to be updated in real time. During the solving process, the numerical values of each variable are affected by historical data and real-time monitoring data, and standardization processing is carried out to ensure the comparability of various data in the model.

[0041] After the preemption decision, some URLLC service requests have obtained priority resource scheduling. At this time, these preemption results need to be fed back to the initial allocation matrix to update the resource allocation in each time slot, frequency band, and spatial unit, and generate a new set of resource allocation decisions to ensure that the actual allocation results meet the latency and reliability requirements of all emergency services. Specifically, each element in the initial slice resource allocation matrix is adjusted. The adjustment operation is based on the decision results of each request in the resource preemption flag vector. If the resource preemption flag vector corresponding to a certain request is 1, it means that the request has successfully preempted resources; at this time, in the time slot, frequency band, and spatial unit corresponding to the corresponding slice, the original allocation value will increase by the proportion of resources that need to be reallocated from the original resources, and the original resource allocation will be reduced accordingly. The path reconstruction instruction set provides operation instructions for network path updates, such as updating the switch flow table, reconfiguring satellite beams, etc. Its instructions will be directly sent to the corresponding devices to complete the actual switching of physical resources.

[0042] After the update operation is completed, the obtained resource allocation sequence is recorded in the form of a matrix as . Each of its elements represents the proportion of resources actually obtained by each slice on the corresponding time slot, frequency band, and spatial unit in the final resource allocation decision. At the same time, each instruction in the path reconstruction instruction set will be sent to the network device through the control platform to adjust the network transmission path in real time to ensure that the service data after preemption can be transmitted along the new path, so as to meet the strict latency requirements.

[0043] At the same time, the update operation adopts a combination of batch processing and real-time feedback. After each solution of the mixed integer linear programming mathematical model, the initial slice resource allocation matrix is immediately updated, and the update results are stored in the database for subsequent query by the scheduling module and monitoring module. During the update process, the remaining resources of each resource unit also need to be re-counted to ensure that the new matrix is consistent with the actual physical resource status.

[0044] Step S4: Update the occupancy rate of the dynamic resource status tensor according to the resource allocation sequence to obtain the ground network resource data, and perform resource fusion processing on the ground network resource data and the received satellite beam capacity data to obtain the space-ground integrated resource pool tensor.

[0045] Since the initial slice resource allocation matrix reflects the preliminary decision results of resource allocation in each dimension, but when performing resource fusion subsequently, a more refined occupancy rate representing the actual resource utilization at each position is required. Therefore, performing occupancy rate extraction processing on the resource allocation sequence can quantify the proportion of resources already allocated at each position, and then calculate the remaining available resources. Specifically, first, from the resource allocation sequence, the proportion of resources allocated to each slice i at time slot t, frequency band f, and spatial unit s is extracted. The extraction process statistically counts the occupancy of each slice on each resource unit according to a predetermined rule, and sums up the proportion of resources occupied by all slices on this resource unit to calculate a comprehensive resource occupancy rate parameter δ(t, f, s). The calculated resource occupancy rate parameter δ(t, f, s) provides quantified occupancy rate information for each resource unit, and will be used as a key parameter to update the dynamic resource status tensor, helping to determine the currently available resources of the terrestrial network.

[0046] Furthermore, the original dynamic resource status tensor R is updated according to the resource occupancy rate parameter δ(t, f, s) to obtain the terrestrial network resource data R ground . Among them, the dynamic resource status tensor R originally described the total theoretical resources in each time slot, frequency band, and spatial unit. However, in practice, some resources have been allocated to each slice for use. Therefore, it is necessary to deduct the occupied resources so that the updated tensor can reflect the actual available terrestrial resource status (i.e., the terrestrial network resource data R ground ). Specifically, a simple proportional deduction method is used to deduct the occupied resource proportion from each resource status value r in the initial dynamic resource status tensor one by one t,f,s to obtain the unoccupied resource amount in each resource unit, thereby obtaining the terrestrial network resource data reflecting the actual available resources.

[0047] At the same time, the satellite beam capacity data obtained from the satellite communication system is processed and virtualized into resource blocks corresponding to the terrestrial network resources. The satellite beam capacity data usually exists in the form of the remaining resource amount of the beam, and this data directly reflects the available communication resources at the satellite end. To achieve integrated space-ground resource scheduling, it is necessary to fuse satellite resources with terrestrial network resources so that the entire network forms a unified resource pool among space, air, and ground. Through virtualization mapping processing, the satellite beam data is converted into a set of virtual resource blocks, which is convenient for subsequent merging with the terrestrial network resource data to form a unified resource pool. Specifically, first, the beam capacity data C sat = [c 1 , ……, c p is obtained from the satellite communication system, where each c j represents the remaining resource amount of the jth satellite beam. The virtualization mapping processing converts each beam capacity data c jMap to time slots, frequency bands, and spatial units of the terrestrial network to form virtual resource blocks. The mapping method depends on preset rules. For example, satellite beam data is allocated to a number of equivalent terrestrial resource units in a certain proportion. The capacity c of each satellite beam j After mapping, it is allocated to the corresponding virtual resource block v j . The virtual resource block not only contains resource quantity information but also retains the time-frequency-space coordinates corresponding to the terrestrial network resources. All virtual resource blocks v j are aggregated into a virtual resource block set V sat ={v 1 , ……, v p}.

[0048] Separate terrestrial resources or satellite resources each have limitations, but through reasonable combination, their respective advantages can be fully utilized to achieve resource complementarity, thereby improving the overall resource utilization rate and service coverage. The obtained terrestrial network resource data is combined with the virtual resource block set to construct a unified resource pool covering the ground and satellites. The construction of the space-ground integrated resource pool tensor enables the unified scheduling of resources in different domains and solves the scheduling difficulties caused by resource heterogeneity. Specifically, first, standardize the data formats of the terrestrial network resource data and the virtual resource block set to ensure consistent representations in the time, frequency, and space dimensions. Then, sum each element of the standardized terrestrial network resource data and the standardized virtual resource block set one by one to obtain the space-ground integrated resource pool tensor.

[0049] During the data processing, the merging operation requires first converting the virtual resource block set into a format with the same dimension as the terrestrial resource data. During the conversion process, determine the position of each virtual resource block in the tensor through the mapping rule and fill in the corresponding resource quantity at that position to form the virtual resource tensor. Subsequently, use mathematical operations to add the values at the corresponding positions of the two tensors to obtain the merged space-ground integrated resource pool tensor. The merging result not only reflects the real-time available resources of the terrestrial network but also includes the additional resources provided by the satellite beams. This resource pool provides a unified and comprehensive resource view for subsequent resource scheduling and dynamic optimization, effectively solving the problem of uneven resource distribution in the network.

[0050] Step S5: Perform traffic prediction processing through a preset long short-term memory network according to the preset historical traffic data and the space-ground integrated resource pool tensor to obtain a bandwidth pre-adjustment strategy.

[0051] It should be understood that historical traffic data records the network traffic conditions in each slice and each time slot over a past period of time. These data usually have different dimensions and data ranges. Directly using the original data may lead to excessive numerical fluctuations during the subsequent training of the deep learning model due to inconsistent numerical scales, affecting the convergence and accuracy of the model. Therefore, normalization processing is an essential preprocessing step. Its main role is to scale the data to the same range, so that each feature is comparable and the training performance of the model is improved. Among them, two common methods are usually adopted for normalization processing. One is min-max normalization, and the other is standardization. The embodiment of this application adopts the standardization method to convert each data point in the historical traffic data according to its mean and standard deviation, so that all data follow a distribution with zero mean and unit variance. The main reason for such processing is that the standardized data is not only convenient for the network model to perform numerical calculations, but also can improve the robustness of the model to outliers. Specifically, the historical traffic data is in vector form, reflecting the traffic statistics of each slice. Calculating the mean and standard deviation of these data is the first step of normalization. After obtaining the mean and standard deviation, each data point in the historical traffic data is subjected to a standardization transformation to obtain the normalized historical traffic input sequence. After the normalization processing is completed, the historical traffic input sequence has been processed with a unified scale, laying a foundation for the input of the subsequent long short-term memory network.

[0052] Network traffic has the characteristics of time series and dynamic changes. The long short-term memory network can effectively capture long-term dependencies and non-linear relationships in time series, predict future traffic trends, and provide a reference basis for subsequent bandwidth adjustment strategies. The normalized historical traffic input sequence and the data of the space-ground integrated resource pool are jointly input into the long short-term memory network to predict the traffic prediction data for a period of time in the future. Among them, the space-ground integrated resource pool tensor reflects the integration of ground and satellite resources and describes the resource distribution within each time slot, frequency band, and spatial unit. While analyzing the historical traffic changes, the long short-term memory network considers the current state of network resources to make a more accurate prediction of future traffic. Specifically, first, the normalized historical traffic input sequence and the space-ground integrated resource pool tensor are appropriately preprocessed. The space-ground integrated resource pool tensor needs to be flattened or reconstructed according to the model input requirements to combine with the time series data. The data preprocessing includes dimension conversion of the space-ground integrated resource pool tensor to match it with the historical traffic data in the time dimension, forming the joint input feature X. The structure adopted by the long short-term memory network includes an input layer, multiple LSTM layers, and an output layer. The input layer receives the joint input feature X, which contains historical traffic information and the current resource state. Inside the long short-term memory network, the gating mechanism controls the forgetting and remembering of information to capture the long-term dependencies of the time series. The output layer then generates the predicted future traffic prediction data. The joint input feature X will first pass through an embedding layer and a convolutional layer to extract local features, and then be input into the multi-layer long short-term memory network. The training process of the network uses the backpropagation algorithm, with the mean square error between the historical data and the real traffic as the loss function, continuously updating the network weights to make the prediction result as close as possible to the actual future traffic.

[0053] Among them, the traffic prediction data is a normalized value and usually needs to be inverse-normalized according to the normalization parameters to restore it to the original data scale. This prediction data not only provides future traffic trends but also reflects the relationship between the current network resource state and traffic changes, providing a scientific basis for subsequent bandwidth adjustment strategies. The processing of the joint input feature during the prediction process makes the traffic prediction not only depend on historical trends but also consider the current resource situation, improving the prediction accuracy and robustness.

[0054] Furthermore, by using the historical average traffic as a benchmark, the deviation degree of the predicted traffic can be quantified, and then the adjustment ratio can be determined, making the resource allocation more in line with future traffic demands, thereby improving service quality and resource utilization. By comparing the traffic prediction data with the average traffic in the historical traffic data, the bandwidth of each slice is dynamically adjusted according to the ratio of the two, thus forming a bandwidth pre-adjustment strategy. The bandwidth pre-adjustment strategy ensures that the slice bandwidth allocation is adjusted in advance before the traffic peak arrives, avoiding network congestion, and releasing redundant bandwidth during the traffic trough, realizing flexible scheduling and efficient utilization of resources. Specifically, first, a calculation method similar to the normalization statistical index is used to calculate the average value of the historical traffic data to obtain the average traffic. Then, the ratio operation is performed on the traffic prediction data and the average traffic within each time slot to determine the bandwidth adjustment factor. This factor is used to correct the original bandwidth, so that when the traffic prediction is high, the bandwidth allocated to the slice increases proportionally; while when the traffic prediction is low, the bandwidth adjustment factor is less than 1, automatically reducing the allocated bandwidth, thereby optimizing resource utilization and service quality. The adjustment process is strictly carried out according to the preset upper limit to ensure that the adjustment amplitude does not exceed the system's tolerance. The adjusted bandwidth constitutes the bandwidth pre-adjustment strategy, and the data structure of the bandwidth pre-adjustment strategy is a vector or matrix, reflecting the bandwidth allocation ratio that each slice should be adjusted in the future period.

[0055] It is difficult for any resource scheduling strategy to foresee all emergencies. Especially when facing critical services such as ultra-reliable low-latency communication, the shortage of local resources is inevitable. At this time, relying solely on internal resource scheduling and prediction can no longer meet the service requirements. External resources must be introduced through cross-domain resource trading to achieve rapid resource replenishment and dynamic balance. In an alternative embodiment, through the automated trading mechanism of blockchain smart contracts, after detecting a local resource gap, cross-operator resource trading is triggered in a timely manner, thereby ensuring that the network scheduling system can obtain sufficient resource guarantees at any time. The specific operations are as follows: After obtaining the bandwidth pre-adjustment strategy, the difference between the locally available resources and the expected demand is quantified using the bandwidth pre-adjustment strategy and the space-ground integrated resource pool tensor, and then the resource gap amount is calculated. By combining the two, it can be clarified whether the current local resources are sufficient and the degree of sufficiency. When the predicted demand exceeds the existing resources, a resource gap is formed. The calculation of this gap amount is crucial for subsequent cross-operator resource transactions. Only by clarifying the gap quantity can the resource replenishment process be initiated to ensure that the network has sufficient resources during the business peak period to meet the requirements of ultra-low latency and high reliability. First, the resource gap amount can be obtained by comparing the total bandwidth demand adjusted according to the bandwidth pre-adjustment strategy with the total resource amount provided by the space-ground integrated resource pool tensor. It should be understood that before calculating the resource gap amount, it is necessary to sum up the resources on all frequency bands and spatial units within time slot t in the space-ground integrated resource pool tensor to obtain the total available resources within that time slot. The calculated resource gap amount is the difference between the predicted demand and the available resources. When the predicted demand is greater than the actual available resources, the resource gap amount is a positive number; otherwise, it is zero.

[0056] When the resource gap amount is greater than zero, construct the cross-operator resource transaction request data. The construction of the cross-operator resource transaction request needs to consider the transaction priority coefficient, which is predefined to distinguish the urgency of different service types (such as URLLC, eMBB, mMTC). The priority coefficient can ensure that when resources are insufficient, higher priority is given to emergency services, so as to obtain the necessary resources preferentially during the cross-domain transaction process. Specifically, when the resource gap amount is greater than zero, according to the transaction priority coefficient α preset for the service type (for example, the highest priority coefficient for URLLC service is 3, for eMBB service is 2, and for mMTC service is 1), construct the transaction request data. Assume that each slice or service module already has a corresponding priority coefficient according to the service requirements. Construct the cross-operator resource transaction request data structure according to the gap amount and the priority coefficient. The keyword fields included in the transaction request data usually include: the resource gap amount, the transaction priority coefficient, and the deadline for resource delivery. Among them, the deadline is determined according to the service requirements and traffic prediction. The constructed cross-operator resource transaction request will be transmitted to the blockchain smart contract module through the data interface and wait for the matching process of subsequent operator quotes.

[0057] Cross-operator resource trading requires reaching an agreement among multiple operators. Only by automating the matching of quotes through smart contracts can efficient, low-latency, and secure resource trading be achieved. Using a preset blockchain smart contract, by performing operator quote matching processing on cross-operator resource trading requests, an encrypted voucher for resource allocation is obtained. Among them, the blockchain smart contract is used to automatically execute trading rules, verify the consistency of quotes and budgets, and generate secure and tamper-proof encrypted vouchers. Through blockchain technology, it can be ensured that trading information is transparent and tamper-proof, and the legitimacy of both trading parties is verified through encryption technology.

[0058] Specifically, when the blockchain smart contract is triggered to execute (i.e., the resource gap volume is greater than zero), first, the budget ceiling is calculated according to the blockchain smart contract with built-in preset trading rules (such as the budget ceiling calculation formula and quote matching conditions). Among them, the calculation of the budget ceiling is usually based on the product calculation of the trading priority coefficient, the resource gap volume, and the preset benchmark resource unit price to obtain the budget ceiling within time slot t (i.e., the maximum allowable payment). Then, according to the quote matching conditions in the blockchain smart contract, traverse the resource quote lists submitted by each participating operator in the blockchain network and verify each quote. When a combination that meets the quote conditions is found (i.e., there is an operator's quote less than or equal to the budget ceiling), an encrypted resource allocation voucher is generated according to the blockchain smart contract. This voucher is generated through asymmetric encryption signatures to ensure the security and tamper-proof nature of trading information. During the process of matching quotes and generating vouchers, the blockchain smart contract records trading information throughout the process and writes it into the blockchain ledger to ensure that the data is publicly transparent and tamper-proof.

[0059] After obtaining the encrypted resource allocation voucher and the resource gap volume, the resources obtained from the outside after the resource trading is completed need to be promptly merged into the internal resource pool to achieve the dynamic balance and optimal utilization of the entire network resources. The encrypted resource allocation voucher and the resource gap volume are used to update the space-ground integrated resource pool tensor, so that the new resource data after the supplementary transaction can be reflected in the overall resource pool, thereby dynamically adjusting the distribution of resources in the network to ensure that after the cross-operator resource trading is completed, the actual available resources can meet future business needs, and at the same time provide the latest data support for subsequent resource scheduling decisions. Specifically, first, according to the obtained resource gap volume and the encrypted resource allocation voucher, the resource volume of the successfully completed supplementary transaction is regarded as newly added resources. Among them, the newly added resources are the resource gap volume or a part of it, depending on the actual matching result. The update operation needs to merge the newly added resources into the original space-ground integrated resource pool tensor to obtain the updated space-ground integrated resource pool tensor.

[0060] In the actual implementation process, data verification and format standardization need to be performed on the newly added resources before the update to ensure that their data formats, dimensions, and dimensions are consistent with those of the ground integrated resource pool tensor before the update. Among them, the resource allocation encryption certificate is used to verify the legality and integrity of this transaction, ensuring that the update operation is based on a true and valid transaction result. The data processing platform will first read the transaction data verified by the encryption certificate from the blockchain ledger and extract the newly added resource quantity. Subsequently, the newly added resource data will be added item by item to the corresponding original resource pool data to complete the update operation. The updated space-ground integrated resource pool tensor will reflect the latest network resource status after cross-operator resource transactions, including both the original ground and satellite resources and the supplementary resources obtained through transactions.

[0061] This application is applied to the field of network resource scheduling technology. By performing three-dimensional joint modeling on heterogeneous network physical resources to obtain a dynamic resource status tensor, making dynamic resource allocation decisions on the dynamic resource status tensor and the service demand matrix through a deep reinforcement learning model to obtain an initial slice resource allocation matrix, performing mixed-integer linear programming on the initial slice resource allocation matrix and the URLLC service request queue through a preemption decision model to obtain a resource allocation sequence, updating the occupancy rate of the dynamic resource status tensor according to the resource allocation sequence to obtain ground network resource data, and performing resource fusion on the ground network resource data and the satellite beam capacity data to obtain a space-ground integrated resource pool tensor, and performing traffic prediction based on historical traffic data and the space-ground integrated resource pool tensor through a long short-term memory network to obtain a bandwidth pre-adjustment strategy. Through the organic integration of multi-dimensional modeling, deep reinforcement learning, mixed-integer linear programming, traffic prediction, and blockchain technology, this application not only improves the intelligent level and real-time response ability of resource scheduling, but also realizes the efficient integration of ground and satellite resources, ensures the low-latency and high-reliability transmission of critical services, and ultimately achieves the goal of improving the overall network resource utilization rate and service quality.

[0062] As Figure 2 shown, it is a functional module diagram of a heterogeneous network intelligent slice resource scheduling device provided by an embodiment of this application.

[0063] In some embodiments, the heterogeneous network intelligent slice resource scheduling device 2 may include multiple functional modules composed of computer program segments. The computer programs of each program segment in the heterogeneous network intelligent slice resource scheduling device 2 can be stored in the memory of the server and executed by at least one processor to execute (see Figure 1 description) the functions of the heterogeneous network intelligent slice resource scheduling method.

[0064] In this embodiment, the heterogeneous network intelligent slice resource scheduling device 2 can be divided into multiple functional modules according to the functions it performs. The functional modules may include: a three-dimensional joint module 21, a resource allocation module 22, a linear programming module 23, a resource fusion module 24, a traffic prediction module 25, and a resource trading module 26. The module referred to in the present invention means a series of computer program segments that can be executed by at least one processor and can complete fixed functions, and are stored in the memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0065] The three-dimensional joint module 21 is used to perform three-dimensional joint modeling processing on the preset heterogeneous network physical resources in the time domain, frequency domain, and spatial domain to obtain a dynamic resource state tensor.

[0066] In an optional implementation manner, the three-dimensional joint module 21 is specifically used for: Performing time slot division processing on the heterogeneous network physical resources according to a preset time window to obtain a time slot set; Performing subcarrier division and priority weighting processing on the heterogeneous network physical resources to obtain a weighted frequency band set; Performing coverage area grid processing on the heterogeneous network physical resources according to the preset base station and satellite coverage areas to obtain a spatial unit set; Performing three-dimensional joint construction processing on the time slot set, the weighted frequency band set, and the spatial unit set to obtain the dynamic resource state tensor.

[0067] The resource allocation module 22 is used to perform dynamic resource allocation decision processing on the dynamic resource state tensor and the received service demand matrix through a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix.

[0068] In an optional implementation manner, the resource allocation module 22 is specifically used for: Performing joint encoding processing on the service demand matrix to obtain a state vector; Performing time-domain-frequency-domain-space three-dimensional resource dynamic allocation processing on the state vector and the dynamic resource state tensor through the deep reinforcement learning model to obtain the initial slice resource allocation matrix.

[0069] The linear programming module 23 is used to perform mixed integer linear programming processing on the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence.

[0070] In an optional implementation manner, the linear programming module 23 is specifically used for: Perform an urgency sorting process on the URLLC service request queue to obtain a priority sequence; Perform a mixed-integer linear function solving process on the priority sequence through the preemption decision model to obtain a resource preemption flag vector and a path reconstruction instruction set; Update the initial slice resource allocation matrix according to the resource preemption flag vector and the path reconstruction instruction set to obtain the resource allocation sequence.

[0071] The resource fusion module 24 is used to update the occupancy rate of the dynamic resource status tensor according to the resource allocation sequence to obtain ground network resource data, and perform resource fusion processing on the ground network resource data and the received satellite beam capacity data to obtain a space-ground integrated resource pool tensor.

[0072] In an optional implementation manner, the resource fusion module 24 is specifically used for: Extract the resource occupancy rate from the resource allocation sequence to obtain a resource occupancy rate parameter; Update the dynamic resource status tensor according to the resource occupancy rate parameter to obtain the ground network resource data; Perform virtualization mapping processing on the satellite beam capacity data to obtain a set of virtual resource blocks; Merge the ground network resource data with the set of virtual resource blocks to obtain the space-ground integrated resource pool tensor.

[0073] The traffic prediction module 25 is used to perform traffic prediction processing through a preset long short-term memory network according to preset historical traffic data and the space-ground integrated resource pool tensor to obtain a bandwidth pre-adjustment strategy.

[0074] In an optional implementation manner, the traffic prediction module 25 is specifically used for: Normalize the historical traffic data to obtain a historical traffic input sequence; Perform traffic prediction processing on the historical traffic input sequence and the space-ground integrated resource pool tensor through the long short-term memory network to obtain traffic prediction data; Perform slice bandwidth dynamic adjustment processing on the traffic prediction data according to the average traffic in the historical traffic data to obtain the bandwidth pre-adjustment strategy.

[0075] In an optional implementation manner, the heterogeneous network intelligent slice resource scheduling device 2 further includes a resource trading module 26, and the resource trading module 26 is specifically used for: Calculate the local resource gap according to the bandwidth pre-adjustment strategy and the space-ground integrated resource pool tensor to obtain the resource gap volume; When the resource gap volume is greater than zero, construct transaction request data for the resource gap volume according to a preset transaction priority coefficient to obtain a cross-operator resource transaction request; Match the operator quotes for the cross-operator resource transaction request according to a preset blockchain smart contract to obtain a resource allocation encryption voucher; Update the space-ground integrated resource pool tensor according to the resource allocation encryption voucher and the resource gap volume.

[0076] It should be understood that the various variations and specific embodiments of the methods provided in the above embodiments are equally applicable to the heterogeneous network intelligent slice resource scheduling device in this embodiment. Through the foregoing detailed description of the heterogeneous network intelligent slice resource scheduling method, those skilled in the art can clearly know the implementation method of the heterogeneous network intelligent slice resource scheduling device in this embodiment. For the sake of brevity of the specification, it will not be elaborated here.

[0077] As Figure 3 shown, it is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0078] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to: a memory 31, at least one processor 32, and at least one communication bus 33.

[0079] Those skilled in the art should understand that Figure 3 the structure of the electronic device 3 shown does not constitute a limitation on the embodiments of the present invention. The electronic device 3 may also include more or fewer other hardware or software than shown, or different component arrangements.

[0080] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, a microprocessor, an application-specific integrated circuit, a programmable gate array, a digital processor, and an embedded device, etc.

[0081] It should be noted that the electronic device 3 is only an example, and other existing or future possible electronic products that can be adapted to the present application should also be included in the protection scope of the present application and are included herein by reference.

[0082] In some embodiments, a computer program is stored in the memory 31, and when the computer program is executed by the at least one processor 32, all or part of the steps in the heterogeneous network intelligent slice resource scheduling method as described above are implemented. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc memories, magnetic disk memories, tape memories, or any other computer-readable medium capable of carrying or storing data. Further, the computer-readable storage medium mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc.

[0083] In some embodiments, the at least one processor 32 is the control core (Control Unit) of the electronic device 3, connecting various components of the entire electronic device 3 through various interfaces and lines. By running or executing the programs or modules stored in the memory 31, and by calling the data stored in the memory 31, various functions of the electronic device 3 are executed and data is processed. For example, when the at least one processor 32 executes the computer program stored in the memory 31, all or part of the steps in the heterogeneous network intelligent slice resource scheduling method described in the embodiments of the present application are implemented; or all or part of the functions of the heterogeneous network intelligent slice resource scheduling device are implemented. The at least one processor 32 may be composed of integrated circuits. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0084] In some embodiments, the at least one communication bus 33 is arranged to enable connection communication between the memory 31, the at least one processor 32, and the like. Although not shown, the electronic device 3 may further include a power source (such as a battery) for powering each component. Preferably, the power source can be logically connected to the at least one processor 32 through a power management device, so as to implement functions such as management of charging, discharging, and power consumption management through the power management device. The power source may further include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or inverter, and a power status indicator. The electronic device 3 may further include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated herein.

[0085] The integrated units implemented in the form of software function modules as described above can be stored in a computer-readable storage medium. The above software function modules are stored in a storage medium and include several instructions for causing an electronic device (which may be a personal computer, an electronic device, or a network device, etc.) or a processor to execute parts of the methods described in various embodiments of the present application.

[0086] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0087] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0088] The above are all preferred embodiments of the present application. Without limiting the protection scope of the present application accordingly, therefore, all equivalent changes made according to the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for scheduling intelligent slice resources in a heterogeneous network, characterized in that: The method comprises: Perform three-dimensional joint modeling of the preset heterogeneous network physical resources in the time domain, frequency domain, and space domain to obtain the dynamic resource state tensor; Performing dynamic resource allocation decision processing on the dynamic resource state tensor and the received business demand matrix through a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix; Performing mixed integer linear programming processing on the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence; Performing occupancy update processing on the dynamic resource state tensor according to the resource allocation sequence to obtain ground network resource data, and performing resource fusion processing on the ground network resource data and the received satellite beam capacity data to obtain a ground-ground integrated resource pool tensor; Traffic prediction processing is performed through a preset long short-term memory network according to preset historical traffic data and the tensor of the integrated space-ground resource pool to obtain a bandwidth pre-adjustment strategy.

2. The method for scheduling heterogeneous network intelligent slice resources according to claim 1, characterized in that: The performing three-dimensional joint modeling processing of the preset heterogeneous network physical resources in the time domain, frequency domain and space domain to obtain a dynamic resource state tensor includes: Performing time slot division processing on the heterogeneous network physical resources according to a preset time window to obtain a time slot set; Performing subcarrier division and priority weighting processing on the heterogeneous network physical resources to obtain a weighted frequency band set; Performing gridding processing on the coverage area of ​​the heterogeneous network physical resources according to the preset base station and satellite coverage areas to obtain a set of spatial units; A three-dimensional joint construction process is performed on the time slot set, the weighted frequency band set, and the spatial unit set to obtain the dynamic resource state tensor.

3. The method for scheduling heterogeneous network intelligent slice resources according to claim 1, characterized in that: The performing dynamic resource allocation decision processing on the dynamic resource state tensor and the received business demand matrix by using a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix includes: Performing joint coding processing on the service demand matrix to obtain a state vector; The state vector and the dynamic resource state tensor are subjected to time-domain-frequency-domain-space three-dimensional resource dynamic allocation processing through the deep reinforcement learning model to obtain the initial slice resource allocation matrix.

4. The method for scheduling heterogeneous network intelligent slice resources according to claim 1, characterized in that: The performing mixed integer linear programming processing on the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence includes: Sorting the URLLC service request queue according to urgency to obtain a priority sequence; Performing mixed integer linear function solving processing on the priority sequence through the preemption decision model to obtain a resource preemption flag vector and a path reconstruction instruction set; The initial slice resource allocation matrix is ​​updated according to the resource preemption flag vector and the path reconstruction instruction set to obtain the resource allocation sequence.

5. The method for scheduling heterogeneous network intelligent slice resources according to claim 1, characterized in that: The performing occupancy updating processing on the dynamic resource state tensor according to the resource allocation sequence to obtain ground network resource data, and performing resource fusion processing on the ground network resource data and the received satellite beam capacity data to obtain a ground-ground integrated resource pool tensor includes: Performing resource occupancy extraction processing on the resource allocation sequence to obtain a resource occupancy parameter; Update the dynamic resource state tensor according to the resource occupancy rate parameter to obtain the ground network resource data; Performing virtualization mapping processing on the satellite beam capacity data to obtain a virtual resource block set; The ground network resource data is combined with the virtual resource block set to obtain the ground-ground integrated resource pool tensor.

6. The method for scheduling heterogeneous network intelligent slice resources according to claim 1, characterized in that: The method of performing traffic prediction processing according to preset historical traffic data and the space-ground integrated resource pool tensor through a preset long short-term memory network to obtain a bandwidth pre-adjustment strategy includes: Normalizing the historical traffic data to obtain a historical traffic input sequence; Performing flow prediction processing on the historical flow input sequence and the space-ground integrated resource pool tensor through the long short-term memory network to obtain flow prediction data; The traffic prediction data is dynamically adjusted for slice bandwidth according to the average traffic in the historical traffic data to obtain the bandwidth pre-adjustment strategy.

7. The method for scheduling heterogeneous network intelligent slice resources according to claim 6, characterized in that: The method further comprises: Perform local resource gap calculation processing according to the bandwidth pre-adjustment strategy and the space-ground integrated resource pool tensor to obtain a resource gap amount; When the resource gap amount is greater than zero, performing transaction request data construction processing on the resource gap amount according to a preset transaction priority coefficient to obtain a cross-operator resource transaction request; Performing operator quotation matching processing on the cross-operator resource transaction request according to a preset blockchain smart contract to obtain a resource allocation encrypted certificate; The space-ground integrated resource pool tensor is updated according to the resource allocation encryption certificate and the resource gap amount.

8. A heterogeneous network intelligent slice resource scheduling device, characterized in that: The device comprises: The three-dimensional joint module is used to perform three-dimensional joint modeling processing on the preset heterogeneous network physical resources in the time domain, frequency domain and space domain to obtain the dynamic resource state tensor; A resource allocation module, used to perform dynamic resource allocation decision processing on the dynamic resource state tensor and the received business demand matrix through a preset deep reinforcement learning model to obtain an initial slice resource allocation matrix; A linear programming module, used to perform mixed integer linear programming processing on the initial slice resource allocation matrix and the received URLLC service request queue through a preset preemption decision model to obtain a resource allocation sequence; A resource fusion module, configured to perform occupancy update processing on the dynamic resource state tensor according to the resource allocation sequence to obtain ground network resource data, and perform resource fusion processing on the ground network resource data and the received satellite beam capacity data to obtain a ground-ground integrated resource pool tensor; The traffic prediction module is used to perform traffic prediction processing according to preset historical traffic data and the tensor of the integrated space-ground resource pool through a preset long short-term memory network to obtain a bandwidth pre-adjustment strategy.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the heterogeneous network intelligent slicing resource scheduling method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the heterogeneous network intelligent slice resource scheduling method according to any one of claims 1 to 7 are implemented.

Images