A 5G-based industrial multi-modal network slice resource allocation system

Abstract

The application relates to the technical field of 5G industrial network resource management, and discloses a 5G-based industrial multi-modal network slice resource allocation system, wherein the system comprises a micro burst detection module, a causal propagation analysis module, a non-uniform elastic reservation module, a resource index calculation module, a Nash bargaining game solution module, a safety constraint checking module, a conditional resource instant recycling module, a feedback calibration module and a long-term adaptive updating module. The system extracts a micro burst interference propagation path through Granger causality test, performs non-uniform elastic reservation based on an interference arrival probability distribution, and utilizes a Nash bargaining game solution to solve cross-slice resource borrowing amount at a time slot granularity, so that sub-period-level resource response of the multi-modal slice is realized.

Description

Technical Field

[0001] This invention relates to the field of 5G industrial network resource management technology, and more specifically, to a 5G-based industrial multimodal network slicing resource allocation system. Background Technology

[0002] In 5G industrial networks, uRLLC slices, eMBB slices, and mMTC slices share the same physical resource pool. These three types of multimodal slices have significantly different requirements for latency, throughput, and connection density. Existing technologies employ two approaches: firstly, statistical cross-correlation methods are used to detect interference relationships between slices. After identifying mMTC micro-burst events, a fixed number of dedicated physical resource blocks are reserved for uRLLC slices as a preventative resource reserve window. Secondly, a periodic scheduling method is used to allocate resources once at the beginning of the scheduling period, without cross-slice resource scheduling within the scheduling period. Both of these resource allocation methods operate independently on the same physical resource pool.

[0003] However, the above scheme has the following drawbacks: The preventative resource reservation window excludes dedicated resource blocks from the schedulable resource pool throughout the reservation period. However, the propagation delay of micro-burst causal interference has variance, and within some time slots of the window, uRLLC slices are not actually affected by the interference, leaving the locked resources idle and wasted. Simultaneously, the periodic scheduling method cannot detect the existence of the preventative reservation window, excluding reserved resources when calculating the available resources for each slice. This leads to a reduction in the overall schedulable resource pool, and multimodal slices cannot receive sufficient resource responses to sub-periodic traffic mutations during micro-burst periods. uRLLC slices experience brief but frequent delay violations within the scheduling cycle, resulting in a corresponding increase in the resource request rejection rate for eMBB and mMTC slices, and a decrease in the efficiency of global multimodal resource allocation. Summary of the Invention

[0004] This invention provides a 5G-based industrial multimodal network slice resource allocation system, which solves the technical problems in related technologies where network slice resource allocation cannot effectively cope with cross-slice interference propagation caused by micro-burst events in mMTC slices, and where fixed reservation strategies result in both resource waste and latency violations.

[0005] This invention discloses a 5G-based industrial multimodal network slicing resource allocation system, including a micro-burst event detection module, a causal propagation analysis module, a non-uniform elastic reservation module, a resource index calculation module, a Nash bargaining game solving module, and a conditional resource instant recovery module; The micro-burst event detection module is configured to perform sliding window change point detection on the resource occupancy sequence of mMTC slices to identify micro-burst events, and extract the start time, duration and peak resource occupancy of the micro-burst events. The causal propagation analysis module is configured to use Granger causality test to analyze the causal propagation path of micro-burst events to uRLLC slices and eMBB slices, generate a micro-burst causal propagation path diagram, and obtain the interference arrival probability distribution of each time slot within the interference impact prediction time window based on historical propagation delay sample statistics. The non-uniform elastic reservation module is configured to allocate the number of reserved resource blocks according to the ratio of the interference arrival probability of each time slot to the maximum probability within the window, and mark the difference resource blocks as conditionally schedulable resources and inject them into the shared resource pool. The resource index calculation module is configured to calculate the resource urgency index and resource surplus index of each slice at the time slot level. The Nash bargaining game solving module is configured to solve the resource borrowing allocation vector between slices based on the resource urgency index and the resource surplus index; The conditional resource instant recovery module is configured to return the borrowed conditional schedulable resources to the uRLLC slice when the queue backlog depth of the uRLLC slice exceeds a preset emergency recovery threshold.

[0006] Furthermore, the causal propagation analysis module uses the start time of the micro-burst event as the starting point. Using the anchor point as a reference, the resource reallocation instruction sequences of uRLLC, eMBB, and mMTC slices within a preset range before and after the anchor point are extracted. Granger causality tests are used to perform hop-by-hop analysis to determine the significance of the causal impact of changes in mMTC slice resource granting on changes in uRLLC and eMBB slice resource granting, as well as the strength and propagation delay values. This generates a micro-burst causal propagation path graph with each slice as a node and directed edges for slice pairs exhibiting significant causal impact. Each directed edge records the causal impact strength and median propagation delay. and the standard deviation of propagation delay .

[0007] Furthermore, the causal propagation analysis module queries the directed edges in the microburst causal propagation path graph that point to the uRLLC and eMBB slices, respectively, to obtain the corresponding median propagation delay. and the standard deviation of propagation delay The starting boundary of the time window for predicting the impact of interference is determined as follows: The end boundary is ,in The preset coverage factor; for the total coverage within the window For each time slot, based on the historical propagation delay samples accumulated in the micro-burst causal propagation path diagram, the frequency of samples falling into the corresponding delay interval of each time slot is counted. The frequency of each time slot is normalized by dividing by the total number of samples to obtain the interference arrival probability of each time slot. ,in For time slot index, This forms the interference arrival probability distribution; the two directed edges pointing from the mMTC slice to the uRLLC slice and from the mMTC slice to the eMBB slice are each calculated independently for the interference arrival probability distribution.

[0008] Furthermore, the non-uniform elastic reservation module reserves space within the window for the first... For each time slot, allocate the number of reserved physical resource blocks according to the following formula. :

[0009] in The total amount of resource blocks originally reserved. For the first The probability of interference arrival in each time slot, This represents the maximum probability of interference arrival within the window. The total number of time slots within the window; the number of schedulable resource blocks available for release under the conditions of each time slot is... The conditionally schedulable resources are kept with an independent identifier in the shared resource pool to distinguish them from regular schedulable resources.

[0010] Furthermore, the resource index calculation module calculates the queue backlog depth for each slice. and the number of remaining available resource blocks After Z-score standardization, the resource urgency index is calculated using the following formula. :

[0011] in For the first The latency-sensitive weighting factor corresponding to each slice type, uRLLC slice Higher than eMBB slices and mMTC slices; when The standardized value is zero and When the standardized value is greater than zero, Take it directly as exceeding the emergency threshold The preset upper limit value; the resource index calculation module calculates the resource surplus index according to the following formula. :

[0012] in For the first The number of remaining available resource blocks for each slice in the current time slot. To maintain the The minimum number of resource blocks required for the current throughput of a slice. This represents the maximum capacity of resource blocks for this slice type; The calculation incorporates the conditionally schedulable resources into the available resource range of each slice.

[0013] Furthermore, the Nash bargaining game solving module will incorporate the resource urgency index. Higher than the preset emergency threshold The slice is marked as the resource borrower, and the resource surplus index is used. Exceeding the preset surplus threshold The slice is marked as the resource lender; define the game utility function. :

[0014] in For the borrower The probability of latency violations decreased for each slice after it obtained borrowed resources. For the lender The throughput loss of each slice due to lending out resources To reduce the weighting coefficient for delay violations, This is the throughput loss weighting coefficient; and Before substitution, mean normalization based on the range is performed separately; the Nash bargaining solution is obtained by solving under the constraint of satisfying the range of resource blocks in each slice. The resource borrowing allocation scheme with the maximum value is obtained, where... For the first The game utility function value corresponding to each participating slice.

[0015] Furthermore, during the borrowing process, the Nash bargaining game solving module distinguishes between regular surplus resources and conditionally schedulable resources based on resource identifiers in the shared resource pool, setting the borrowing priority of conditionally schedulable resources to be lower than that of regular surplus resources. When the lender's regular surplus resources are sufficient to meet the borrowing amount specified by the resource borrowing allocation vector, only regular surplus resources are borrowed. When regular surplus resources are insufficient to meet the borrowing amount, conditionally schedulable resources are used to participate in cross-slice borrowing based on the difference. The system also includes a security constraint verification module, configured to calculate the number of remaining resource blocks after the borrowing for each lender's slice. ,verify Whether it is valid, among which For the first The number of remaining available resource blocks for each lender slice. For the amount lent out, This is the minimum service guarantee threshold for this slice type; if the constraint check fails, the loan amount will be reduced to [a certain value]. Any shortfall in lending resulting from the reduction shall be borne by all borrowers in equal proportion to their original allocation.

[0016] Furthermore, the system also includes a feedback calibration module, configured to, after the interference impact prediction time window ends, statistically analyze the prediction deviation between the actual interference arrival time of each time slot within the window and the predicted probability peak time slot by the interference arrival probability distribution, and use the prediction deviation to update the median propagation delay of the corresponding directed edge in the micro-burst causal propagation path graph. and the standard deviation of propagation delay The system calculates the net resource utilization benefit of conditionally borrowed resources by statistically analyzing the borrowing rate, recovery rate, and recovery response latency of conditionally schedulable resources during the window period. The net resource utilization benefit is the percentage of time slots in which conditionally schedulable resources are effectively borrowed and utilized minus the percentage of time slots incurred due to scheduling overhead caused by recovery operations. When the net resource utilization benefit is higher than a preset benefit threshold, the maximum release ratio of conditionally schedulable resources is increased; when the net resource utilization benefit is lower than the preset benefit threshold, the maximum release ratio is reduced.

[0017] Furthermore, the system also includes a long-term adaptive update module, configured to update the latency-sensitive weighting factor of each slice type at the end of each scheduling cycle based on the accumulated borrowing revenue records during that cycle using an exponentially weighted moving average method. Weighting coefficients in the game utility function , The borrowing revenue record includes the changes in latency violation rate and throughput of each slice after borrowing, and the updated parameters take effect in subsequent scheduling cycles.

[0018] This invention discloses a 5G-based industrial multimodal network slice resource allocation method for executing the aforementioned 5G-based industrial multimodal network slice resource allocation system. The method includes: collecting resource authorization records and real-time performance indicators of uRLLC slices, eMBB slices, and mMTC slices in each transmission time interval through a micro-burst event detection module; performing sliding window change point detection on the resource occupancy rate sequence of mMTC slices to identify micro-burst events; and extracting the start time, duration, and peak resource occupancy of the micro-burst events. The causal propagation analysis module uses Granger causality test to analyze the causal propagation path of micro-burst events to uRLLC and eMBB slices, generates a micro-burst causal propagation path diagram, and obtains the interference arrival probability distribution of each time slot within the interference impact prediction time window based on historical propagation delay sample statistics. The non-uniform elastic reservation module allocates the number of reserved resource blocks according to the ratio of the interference arrival probability in each time slot to the maximum probability within the window, and marks the difference resource blocks as conditionally schedulable resources and injects them into the shared resource pool. The resource urgency index and resource surplus index of each slice are calculated at the time slot level using the resource index calculation module. The Nash bargaining game solution module solves the resource borrowing allocation vector between slices based on the resource urgency index and resource surplus index, and performs cross-slice resource borrowing. When the queue backlog depth of a uRLLC slice exceeds a preset emergency recovery threshold, the conditionally schedulable resources that have been borrowed are returned to the uRLLC slice via the conditional resource instant recovery module.

[0019] The beneficial effects of this invention are as follows: Firstly, the causal propagation analysis module uses the Granger causality test to capture the causal propagation path and delay characteristics of mMTC micro-burst events to uRLLC and eMBB slices at the transmission time interval level. It can distinguish between real causal interference relationships and temporally accidental spurious associations. The micro-burst causal propagation path map provides a directional and delay-quantified description of interference propagation, providing a more accurate positioning basis for the temporal allocation of reserved resources.

[0020] Secondly, the non-uniform elastic reservation module transforms the fixed resource locking within the preventive reservation window into non-uniform elastic reservation based on the interference arrival probability distribution. It reduces the reservation amount for time slots with low interference arrival probability and releases the difference as conditionally schedulable resources into the multimodal slice shared resource pool. While maintaining the protection strength of high-probability time slots, it releases idle resources of low-probability time slots, thereby reducing the resource waste caused by fixed reservation locking all time slots equally.

[0021] Third, the resource index calculation module incorporates conditionally schedulable resources into the resource surplus index calculation of each slice, and the Nash bargaining game solving module performs cross-slice resource borrowing at the time slot granularity, so that the schedulable resource pool will not shrink as a whole due to preventive reservation during micro-burst periods. Sub-periodic traffic mutations in multimodal slices can obtain timely resource responses, improving the problem of insufficient sub-periodic mutation response caused by the inability to perform cross-slice resource scheduling within the scheduling cycle during periodic scheduling.

[0022] Fourth, when the queue backlog depth of the uRLLC slice exceeds the preset emergency recovery threshold, the conditional resource real-time recovery module will return the borrowed conditional schedulable resources to the uRLLC slice at the next time slot boundary, thus avoiding the uRLLC slice resource security risks introduced after the non-uniform elastic reservation and release of conditional schedulable resources.

[0023] Fifth, the feedback calibration module updates the propagation delay distribution parameters at the window level using calibration deviation, while the long-term adaptive update module adjusts the delay-sensitive weight factor and the game utility function weighting coefficient at the scheduling cycle level using the exponential weighted moving average method. The synergistic effect of parameter updates at the two levels enables the system to gradually improve the accuracy of interference prediction and the efficiency of cross-slice resource utilization during continuous operation. Attached Figure Description

[0024] Figure 1 This is a flowchart of a 5G-based industrial multimodal network slice resource allocation method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall architecture of the multimodal network slice resource scheduling system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the time series of mMTC slice resource occupancy rate and the interval of micro-burst events provided in the embodiments of the present invention; Figure 4 This is a schematic diagram comparing the arrival probability distributions of two directed edges under interference, provided in an embodiment of the present invention. Figure 5 This is a schematic diagram of the non-uniform elastic reservation results of uRLLC slices and eMBB slices provided in the embodiments of the present invention; Figure 6 This is a schematic diagram of the resource urgency index and resource surplus index of each slice in TTI 46 time slot provided in the embodiment of the present invention; Figure 7 This is a schematic diagram of the Nash bargaining game resource borrowing path and physical resource block remapping provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the uRLLC slice queue backlog depth and conditional resource reclamation triggering process provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the parameter iteration results of the long-term adaptive update module provided in the embodiment of the present invention. Detailed Implementation

[0025] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0026] At least one embodiment of the present invention discloses a 5G-based industrial multimodal network slicing resource allocation system, the method of which is described below. Figure 1 The system includes: a micro-emergency event detection module, a causal propagation analysis module, a non-uniform elastic reservation module, a resource index calculation module, a Nash bargaining game solution module, a security constraint verification module, a conditional resource instant recovery module, a feedback calibration module, and a long-term adaptive update module.

[0027] The micro-burst event detection module is configured to collect multimodal slice operation data and identify micro-burst events in mMTC slices.

[0028] The micro-burst event detection module collects resource grant records and real-time performance metrics for uRLLC slices, eMBB slices, and mMTC slices in 5G industrial networks within each transmission time interval (TTI). Resource grant records include the number of physical resource blocks granted to each slice within each TTI, their frequency domain location, and modulation / coding scheme. Real-time performance metrics include the queue backlog depth, single-packet latency, and throughput for each slice.

[0029] The change point detection submodule performs sliding window change point detection on the resource occupancy rate sequence of the mMTC slice. The input of the sliding window change point detection is the resource occupancy rate time series of the mMTC slice with a sampling interval of TTI and the length of the sliding window. The output is a set of change point locations. Based on the set of change point locations, the change point detection submodule identifies intervals in which the resource occupancy rate exceeds a preset multiple of the sliding window mean within a certain number of consecutive TTIs, and marks these intervals as micro-burst events.

[0030] The event parameter extraction submodule extracts the start time for each micro-burst event. Duration and peak resource usage ,in The TTI moment corresponding to the first turning point of a micro-emergency event. This refers to the number of time periods (TTIs) covered by a micro-burst event from its initial change point to its final change point. This represents the maximum resource occupancy rate within the interval of a minor emergency.

[0031] The causal propagation analysis module is configured to analyze the causal propagation path of micro-burst events on cross-slice resource authorization and generate an interference arrival probability distribution.

[0032] The causal propagation analysis module uses the initiation time of micro-emergence events as an example. Using the anchor point as the reference, extract the resource reallocation instruction sequence of uRLLC slices, eMBB slices, and mMTC slices within a preset range before and after the anchor point.

[0033] The causality test submodule uses the Granger causality test method to analyze the causal impact of changes in resource granting for the mMTC slice on changes in resource granting for the uRLLC and eMBB slices, hop-by-hop. The Granger causality test takes as input the time series of resource granting changes for the mMTC slice and the target slice (uRLLC or eMBB slice) within a preset range before and after the anchor point, along with the maximum lag order. The output is the significance of the causal impact between each slice pair, the causal impact strength value, and the propagation delay value.

[0034] The path graph generation submodule generates a micro-burst causal propagation path graph based on the Granger causality test results. The micro-burst causal propagation path graph uses uRLLC, eMBB, and mMTC slices as nodes, and slice pairs with significant causal influence as directed edges. Each directed edge records the causal influence strength and the median propagation delay. and the standard deviation of propagation delay .

[0035] The interference probability calculation submodule queries the directed edges in the micro-burst causal propagation path graph that point to the uRLLC and eMBB slices, respectively, and obtains the corresponding median propagation delay. and the standard deviation of propagation delay The initial boundary of the prediction time window for the impact of disturbances is... The ending boundary is ,in To preset the coverage coefficient, This refers to the starting point of a minor emergency. The median propagation delay, Let $\mathbf{x}$ be the standard deviation of the propagation delay. For the total $\mathb Each time slot Given the total number of time slots within the window, based on historical propagation delay samples accumulated in the micro-burst causal propagation path graph, the frequency of samples falling into the corresponding delay interval of each time slot is counted. The frequency of each time slot is normalized by dividing by the total number of samples to obtain the interference arrival probability of each time slot. ( ),in The time slot index forms the probability distribution of interference arrival.

[0036] It should be noted that the above interference arrival probability distributions are calculated independently for the two directed edges pointing from the mMTC slice to the uRLLC slice and from the mMTC slice to the eMBB slice, respectively, and each of the two directed edges has an independent median propagation delay. Standard deviation of propagation delay Given the historical propagation delay sample set, the interference impact prediction time window range corresponding to the two directed edges and the interference arrival probability distribution of each time slot within the window are independent of each other. The subsequent non-uniform elastic reservation module performs non-uniform resource reservation for uRLLC slices and eMBB slices respectively using their respective interference arrival probability distributions.

[0037] The non-uniform elastic reservation module is configured to non-uniformly allocate reserved resources based on the probability distribution of interference arrival and release conditionally schedulable resources.

[0038] The non-uniform elastic reservation module receives the interference arrival probability distributions of mMTC slices pointing to uRLLC slices and mMTC slices pointing to eMBB slices, as well as the total amount of original fixed reserved resource blocks corresponding to each target slice. The preventative resource reservation is non-uniformly distributed across time slots within each window. The following explanation of the non-uniform distribution process uses a single target slice as an example; uRLLC slices and eMBB slices each execute the same process independently.

[0039] The non-uniform distribution submodule applies to the first element within the window. Each time slot, according to the probability of interference arrival The number of reserved physical resource blocks allocated relative to the proportion of the highest probability within the window. :

[0040] in For the first The number of reserved physical resource blocks per time slot The total amount of resource blocks originally reserved. For the first The probability of interference arrival in each time slot, This represents the maximum probability of interference arrival within the window. This represents the total number of time slots within the window. This is a time slot index. Because... ,therefore This always holds true. The time slot with the highest probability of interference arrival receives an equal value. The full reserve is obtained, while the reserve is proportionally reduced for time slots with a lower probability of interference.

[0041] The conditional resource release submodule calculates the difference in the number of resource blocks released in each time slot:

[0042] in For the first The number of schedulable resource blocks that can be released under the conditions of each time slot. The total amount of resource blocks originally reserved. For the first The number of reserved physical resource blocks for each time slot. The conditional resource release submodule marks the difference in resource blocks as conditionally schedulable resources and injects them into the multimodal slice shared resource pool. Conditionally schedulable resources maintain an independent identifier in the multimodal slice shared resource pool to distinguish them from regular schedulable resources. This independent identifier is used for priority control of borrowing and immediate reclamation operations of conditionally schedulable resources in subsequent modules.

[0043] The resource index calculation module is configured to calculate the resource urgency index and resource surplus index of each multimodal slice at the time slot level.

[0044] The resource index calculation module uses time slots as the smallest calculation granularity, calculating the resource urgency index and resource surplus index for uRLLC slices, eMBB slices, and mMTC slices respectively. Before calculation, the queue backlog depth is assessed. and the number of remaining available resource blocks Z-score standardization was performed separately to eliminate the impact of the difference in the dimensions of the two types of indicators on the calculation of the resource urgency index; the formula for the resource surplus index was adjusted accordingly. , and All data are based on the number of physical resource blocks as a unified unit, requiring no additional normalization processing.

[0045] The resource urgency calculation submodule calculates the resource urgency index for each slice. :

[0046] in For the first The queue backlog depth of each slice in the current time slot after Z-score normalization. For the first The number of remaining available resource blocks in the current time slot after Z-score normalization for each slice. For the first The latency-sensitive weight factor corresponding to each slice type The value of covers three types of slices: uRLLC slices, eMBB slices, and mMTC slices. uRLLC slices carry motion control commands, and their time-delay sensitive weighting factor... Higher than eMBB slices and mMTC slices. When the... Number of remaining available resource blocks per slice The normalized value is zero and the queue backlog depth is zero. When the standardized value is greater than zero, the resource urgency index of the slice is... Take it directly as exceeding the emergency threshold A preset upper limit is set to avoid calculation failures caused by a denominator of zero and to ensure that the slice is marked as a resource borrower.

[0047] The resource surplus calculation submodule calculates the resource surplus index for each slice. :

[0048] in For the first The number of remaining available resource blocks for each slice in the current time slot. To maintain the The minimum number of resource blocks required for the current throughput of a slice. This represents the maximum capacity of resource blocks for this slice type. The value range covers three types of slices: uRLLC slices, eMBB slices, and mMTC slices. The calculation incorporates resource blocks marked as conditionally schedulable resources in the multimodal slice shared resource pool into the available resource range of each slice, so that the release of conditionally schedulable resources is directly reflected in the increase of the redundancy of each slice.

[0049] The Nash bargaining game solution module is configured to solve the resource borrowing allocation vector between multimodal slices based on the resource urgency index and the resource surplus index.

[0050] The Nash bargaining game solving module calculates the resource urgency index for each multimodal slice. and resource surplus index Input a distributed Nash bargaining game solver.

[0051] The role-based submodule will include a resource urgency index. Higher than the preset emergency threshold The slice is marked as the resource borrower, and the resource surplus index is used. Exceeding the preset surplus threshold The slice is marked as the resource lender.

[0052] The utility function submodule defines the game utility function. :

[0053] in For the borrower The probability of latency violations decreased for each slice after it obtained borrowed resources. For the lender The throughput loss of each slice due to lending out resources To reduce the weighting coefficient for delay violations, This is the throughput loss weighting coefficient. For the borrower's slice index, This is the slice index for the lender. and Since their dimensions are different, mean normalization based on the range is performed on both before substituting them into the game utility function, thus unifying them and mapping them to a uniform mean. The weighted calculation is performed after the interval is calculated. The delay violation probability is given by the difference between the borrower's slice before and after the borrowing, i.e., the delay violation probability before borrowing minus the delay violation probability after borrowing. The delay violation probability in each state is determined by the current slot queue backlog depth. The ratio of the corresponding available resource blocks to the preset latency violation probability mapping table is used to obtain the value. For the borrower The queue backlog depth of each slice in the current time slot after Z-score normalization; The throughput difference of the lending slice before and after the borrowing is given by the difference between the throughput before and after the borrowing. The throughput in each state is calculated by the product of the number of remaining available resource blocks in the current time slot and the throughput of a single resource block corresponding to the current modulation and coding scheme of the slice.

[0054] The Nash bargaining solution submodule is based on the game utility function. Solve the Nash bargaining problem. The distributed Nash bargaining game solver takes as input the game utility function value of each slice and the range of resource blocks that each slice can lend or needs to borrow. The output is a vector of resource borrowing amounts among the multimodal slices. The Nash bargaining solution is obtained by maximizing the product of the game utility function values ​​of all participants; that is, under the constraint of the range of resource blocks in each slice, the solution is found to maximize the product of the game utility function values ​​of all participants. The resource borrowing allocation scheme that takes the maximum value, where For the first The game utility function value corresponding to each participating slice. To participate in the slice index.

[0055] During the borrowing process, the priority constraint submodule distinguishes between regular surplus resources and conditionally schedulable resources based on the resource identifiers in the multimodal slice shared resource pool, setting the borrowing priority of conditionally schedulable resources to be lower than that of regular surplus resources. When the lender's regular surplus resources are sufficient to meet the borrowing amount specified by the resource borrowing allocation vector, only regular surplus resources are borrowed; when regular surplus resources are insufficient to meet the borrowing amount, conditionally schedulable resources are used to participate in cross-slice borrowing for the difference.

[0056] The security constraint verification module is configured to apply security constraints to the resource borrowing allocation vector and generate physical resource block remapping instructions.

[0057] The security constraint verification module performs security constraint verification on each lender slice in the resource borrowing allocation vector.

[0058] The constraint verification submodule calculates the number of remaining resource blocks for each lender slice after the borrowing process. :

[0059] in For the first The number of remaining available resource blocks for each lender slice in the current time slot. Assign the first element to the resource borrowing vector. The amount of slices lent out by each lender. For the first The number of remaining resource blocks for each lender slice after the borrowing process. For the lender's slice index. Constraint verification submodule verification. Does it meet the requirements? ,in This is the minimum service guarantee threshold for this slice type.

[0060] When the constraint check fails, the reduction normalization submodule reduces the lending amount of the lender slice that does not meet the constraint to [amount missing]. ,in This is the reduced lending amount. For the first The number of remaining available resource blocks for each lender slice in the current time slot. This is set as the minimum service guarantee threshold for this slice type, ensuring that the number of remaining resource blocks after the slice is borrowed is exactly equal to the minimum service guarantee threshold. Any shortfall in borrowing resulting from the reduction is shared proportionally by all borrowers according to the original allocation ratio. The reduced resource borrowing allocation vector is then renormalized to maintain the original allocation ratio among borrowers.

[0061] The remapping instruction generation submodule converts the verified resource borrowing allocation vector into a physical resource block remapping instruction, specifying the physical resource block number released by each borrower slice and the corresponding receiving binding relationship of the borrower slice, and performs the resource block remapping operation in the current time slot.

[0062] The conditional resource real-time reclamation module is configured to monitor the uRLLC slice status and reclaim the borrowed conditional schedulable resources when interference actually arrives.

[0063] The conditional resource real-time recycling module continues to run after conditionally schedulable resources are borrowed, and monitors the queue backlog depth and single packet latency of uRLLC slices in the corresponding time slot in real time.

[0064] The monitoring and triggering submodule continuously compares the queue backlog depth of the current time slot of the uRLLC slice with a preset emergency recovery threshold. When the queue backlog depth exceeds the preset emergency recovery threshold, it is determined that the interference has actually arrived, triggering the immediate recovery process.

[0065] The resource return submodule sends a resource reclamation command to the slice currently borrowing the conditionally schedulable resource, and returns the conditionally schedulable resource to the uRLLC slice at the next time slot boundary, restoring its reserved resource guarantee for that time slot. After the reclamation is completed, the resource return submodule records the trigger time and reclamation response delay of this reclamation operation for use by the feedback calibration module.

[0066] The feedback calibration module is configured to calibrate the interference prediction parameters and adjust the upper limit of the conditionally schedulable resource ratio after the interference impact prediction time window ends.

[0067] The feedback calibration module performs dual-channel feedback calibration after the interference impact prediction time window ends.

[0068] The calibration deviation calculation submodule counts the actual arrival time of interference in each time slot within the window, compares the actual arrival time of interference with the predicted peak time slot of the interference arrival probability distribution, calculates the prediction deviation of each time slot, and summarizes it into the calibration deviation of the entire window.

[0069] The net revenue calculation submodule calculates the borrowing rate, recovery rate, and recovery response latency of conditionally schedulable resources during the window period. The net resource utilization revenue of conditional borrowing is defined as the percentage of time slots in which conditionally schedulable resources are effectively borrowed and utilized, minus the percentage of time slots incurred due to scheduling overhead caused by recovery operations.

[0070] The parameter update submodule updates the median propagation delay of the corresponding directed edge in the microburst causal propagation path graph using the calibration bias. and the standard deviation of propagation delay This makes the probability distribution of interference arrival in subsequent micro-incidents more closely resemble the actual propagation characteristics. Simultaneously, the upper limit of the conditionally schedulable resource ratio is adjusted using net resource utilization benefits: when net resource utilization benefits exceed a preset benefit threshold, the maximum release ratio of conditionally schedulable resources in the non-uniform elastic reservation module is increased; when net resource utilization benefits fall below the preset benefit threshold, the maximum release ratio is reduced.

[0071] The long-term adaptive update module is configured to update resource allocation strategy parameters based on accumulated borrowing revenue records.

[0072] At the end of each scheduling cycle, the long-term adaptive update module updates the latency-sensitive weighting factor for each multimodal slice type using an exponentially weighted moving average method, based on the accumulated borrowing revenue records during that cycle. Weighting coefficients in the game utility function , The exponentially weighted moving average method takes as input the average borrowing revenue metric for each slice in the current period and the parameter value from the previous period, and outputs the updated parameter value. The borrowing revenue metric includes the changes in latency violation rate and throughput for each slice after the borrowing. The updated latency-sensitive weighting factor... and weighting coefficients , It will take effect in subsequent scheduling cycles, among which For the first The latency-sensitive weight factor corresponding to each slice type To reduce the weighting coefficient for delay violations, This is the throughput loss weighting coefficient. The value range covers three types of slices: uRLLC slices, eMBB slices, and mMTC slices, enabling the multimodal resource allocation strategy to adapt to the long-term evolution of industrial production line flow patterns.

[0073] This invention utilizes the Granger causality test in its causal propagation analysis module to capture the causal propagation path and delay characteristics of mMTC micro-burst events to uRLLC and eMBB slices at the TTI level. It can distinguish between genuine causal interference relationships and temporally coincidental spurious associations. Compared with the associations extracted by statistical cross-correlation methods, the micro-burst causal propagation path map provides a directional and delay-quantified description of interference propagation, providing a more accurate positioning basis for subsequent time-domain allocation of reserved resources.

[0074] This invention transforms the fixed resource locking within a preventative reservation window into a non-uniform elastic reservation based on the interference arrival probability distribution through a non-uniform elastic reservation module. Since the propagation delay of micro-burst causal interference has variance, the probability of each time slot being affected by interference differs within the window. Non-uniform elastic reservation reduces the reservation amount for time slots with lower interference arrival probabilities and releases the difference as conditionally schedulable resources into the multimodal slice shared resource pool. This maintains the protection strength of high-probability time slots while releasing idle resources in low-probability time slots, thus solving the resource waste caused by fixed reservation locking all time slots equally.

[0075] This invention incorporates conditionally schedulable resources into the resource surplus index calculation of each slice through a resource index calculation module, and performs cross-slice resource borrowing at the time slot granularity through a Nash bargaining game solving module. This prevents the schedulable resource pool from shrinking as a whole due to preventative reservations during micro-burst periods. The joint calculation of the resource urgency index and the resource surplus index quantifies the supply and demand status of each slice in real time in each time slot. Based on this, the distributed Nash bargaining game solver obtains an equilibrium allocation between the latency gain of the borrower and the throughput cost of the lender, enabling timely resource responses to sub-periodic traffic mutations in multimodal slices. This solves the problem of insufficient sub-periodic mutation response caused by the inability to perform cross-slice resource scheduling within the scheduling cycle of periodic scheduling.

[0076] This invention utilizes a conditional resource immediate recovery module to return borrowed conditional schedulable resources to the uRLLC slice at the next time slot boundary when the queue backlog depth of the uRLLC slice exceeds a preset emergency recovery threshold. This avoids the uRLLC slice resource security risks introduced by non-uniform elastic reservation and release of conditional schedulable resources. A feedback calibration module updates propagation delay distribution parameters at the window level using calibration bias to improve interference prediction accuracy. A long-term adaptive update module adjusts the delay-sensitive weight factor and game utility function weighting coefficient at the scheduling cycle level using an exponentially weighted moving average method. The synergistic effect of these two levels of parameter updates gradually improves the accuracy of interference prediction and the efficiency of cross-slice borrowing resource utilization during continuous operation.

[0077] See Figures 2-9 The following application examples are proposed: A 5G smart manufacturing production line was put into operation in Cycle A of 20XX. The production line is configured with three types of multimodal network slices sharing the same physical resource pool, with a total of 120 schedulable physical resource blocks. The uRLLC slice carries the motion control commands of the robotic arm, with a latency requirement of no more than 1ms; the eMBB slice carries the high-definition video stream from the production line's visual quality inspection cameras; and the mMTC slice carries the periodic reported data from distributed temperature, vibration, and pressure sensors on the production line. In a certain TTI sequence, the mMTC slice triggered a micro-burst event due to a large number of sensors simultaneously reporting data, and the various modules of the system intervened in sequence to handle it.

[0078] Execution process of the micro-burst event detection module: The change point detection submodule performs sliding window change point detection on the resource utilization time series of the mMTC slice, with a window length of 8 TTIs. It detects a significant jump in resource utilization at TTI 42, followed by a return to normal at TTI 47, identifying a micro-burst event. The event parameter extraction submodule extracts the core parameters of this micro-burst event: the start time. (TTI number), duration (Number of TTIs), Peak resource usage .

[0079] Table 1. Parameters of micro-emergencies and their relationships with different slices. runtime status

[0080] The execution process of the causal propagation analysis module: The causality test submodule is based on Using this as an anchor point, resource grant change sequences for the three types of slices were extracted within 16 TTIs before and after the anchor point. With a maximum lag order of 5, Granger causality tests were performed on the mMTC slice and the uRLLC and eMBB slices, respectively. The test results showed that the mMTC slice had a significant causal effect on the uRLLC slice (p < 0.01), with a causal effect strength of 0.74 and a median propagation delay of [missing value]. (TTI), standard deviation of propagation delay (TTI); The causal effect of the mMTC slice on the eMBB slice is significant (p < 0.05), with a causal effect strength of 0.51 and a median propagation delay. (TTI), standard deviation of propagation delay (TTI). The path graph generation submodule generates a micro-burst causal propagation path graph based on this, and records the parameters of the two directed edges.

[0081] The interference probability calculation submodule is based on To determine the coverage factor, calculate the predicted time window for the interference impact corresponding to each of the two directed edges. For the directed edge from mMTC to uRLLC, the initial boundary of the window is... Take TTI 46, and the end boundary is... Take TTI 48, total within the window There are three time slots (TTI 46, 47, and 48). After normalization based on historical propagation delay samples, the interference arrival probabilities for each time slot are as follows: , , For the directed edge from mMTC to eMBB, the initial window boundary is... Take TTI 47, and the end boundary is... Take TTI51, the total number of [items] in the window is [number]. The interference arrival probability for each time slot (TTI 47 to 51) is shown in Table 2.

[0082] Table 2. Interference arrival probability distribution of two directed edges

[0083] Execution process of the non-uniform elastic reservation module: Total original fixed reserved resource blocks Both uRLLC and eMBB slices are set to 10. Taking uRLLC slices as an example, the maximum interference arrival probability within the window is... (TTI 47), the reserved amount for each time slot is calculated according to the formula. The calculation results for TTI 47 time slot are as follows: (number); The calculation results for TTI 46 time slots are: (number); The calculation results for TTI 48 time slots are: (Number). The conditional resource release submodule calculates the number of conditionally schedulable resource blocks to be released in each time slot. It is then injected into the multimodal slice shared resource pool, maintaining an independent identifier. eMBB slices are processed using the same procedure, with the maximum interference arrival probability within their window being [the specified value]. (TTI 49).

[0084] Table 3 Results of Non-uniform Elastic Retention in uRLLC and eMBB Slices

[0085] Taking TTI 46 as an example, the uRLLC slice releases 5 conditionally schedulable resource blocks and injects them into the shared resource pool, increasing the total number of conditionally schedulable resources in the shared resource pool of this time slot by 5. When calculating resource redundancy, all slices can include these 5 resource blocks in the available range.

[0086] Execution process of the resource index calculation module: Taking TTI 46 slot as an example, the resource urgency index and resource surplus index are calculated after Z-score standardization for the three types of slices. Standardization is performed based on the historical statistical mean and standard deviation of the corresponding indicators for the three types of slices within the scheduling cycle. The latency-sensitive weighting factor for uRLLC slices is also included. eMBB slices mMTC slice Minimum number of resource blocks required to maintain current throughput. and resource block capacity limit As shown in Table 4, the number of available resource blocks in the TTI 46-slot uRLLC slice after incorporating conditionally schedulable resources is... One, resource urgency index and resource surplus index The calculation results are shown in Table 4. Taking uRLLC slices as an example, after standardization... , ,but Exceeding the emergency threshold It is marked as a resource borrower.

[0087] Table 4. Calculation results of resource index for each slice in TTI 46 slot.

[0088] Execution process of the Nash bargaining game solving module: The role division submodule confirms that the uRLLC slice is the borrower ( eMBB slices and mMTC slices are for the lender ( The values ​​are 0.27 and 0.56 respectively, both higher than the surplus threshold. The utility function submodule obtained the latency violation probability of 0.38 before borrowing the uRLLC slice by looking up a table. After borrowing 8 resource blocks, it dropped to 0.11. Throughput loss after eMBB slice borrows 4 resource blocks After normalization, the throughput loss after borrowing 4 resource blocks in an mMTC slice is 0.06 after normalization, which is 0.09. Let... , After substituting the game utility function values ​​of each participant, the Nash bargaining solution submodule maximizes... The resource borrowing allocation vector is obtained as follows: The uRLLC slice borrows a total of 8 resource blocks, of which 4 are borrowed from the eMBB slice (all are regular surplus resources) and 4 are borrowed from the mMTC slice (including 2 conditionally schedulable resources). The priority constraint submodule confirms that the eMBB slice has sufficient regular surplus resources and borrows all of them; the mMTC slice has only 2 regular surplus resources remaining, with the 2 shortfall being the activation of conditionally schedulable resources.

[0089] The execution process of the security constraint verification module: The constraint verification submodule performs eMBB slice calculations. (Number), Minimum Service Guarantee Threshold (indivual), Verification passed; mMTC slice calculation (indivual), (indivual), Verification passed. Both borrowing paths passed constraint verification. The remapping instruction generation submodule generated physical resource block remapping instructions, specifying that four resource blocks with release numbers PRB-31 to PRB-34 of the eMBB slice are bound to the uRLLC slice, and four resource blocks with release numbers PRB-71 to PRB-74 of the mMTC slice (of which PRB-73 and PRB-74 are conditionally schedulable resources) are bound to the uRLLC slice. The remapping is performed within TTI 46.

[0090] Table 5 Security Constraint Verification and Remapping Instructions

[0091] The execution process of the conditional resource real-time recycling module: After conditionally schedulable resources PRB-73 and PRB-74 are loaned out, the conditional resource instant reclamation module continuously monitors the queue backlog depth of the uRLLC slice during TTI 46 to TTI 48. In TTI 47, the uRLLC slice queue backlog depth exceeds the preset emergency reclamation threshold (original value reaches 31 packets). The monitoring trigger submodule determines that interference has actually arrived and triggers the instant reclamation process. The resource return submodule sends a resource reclamation command to the mMTC slice, returning PRB-73 and PRB-74 to the uRLLC slice at the start boundary of TTI 48, restoring its reserved resource guarantee in TTI 48. The response delay for this reclamation is one TTI. The trigger time (TTI 47) and the reclamation response delay are recorded for use by the feedback calibration module.

[0092] The execution process of the feedback calibration module and the long-term adaptive update module: After the interference impact prediction time window (TTI 46 to TTI 48) ends, the actual interference arrival time within the statistical window of the calibration deviation calculation submodule is concentrated in TTI 47, and the probability peak time slot of the interference arrival probability distribution is also TTI 47. The prediction deviation is 0 TTIs, indicating that the prediction is accurate. The net benefit calculation submodule statistical conditions show that the schedulable resource lending rate is 2 / 3 (lending in both TTI 46 and TTI 47, with recovery triggered in TTI 47), the recovery rate is 1 / 2, and the recovery response delay is 1 TTI. The net resource utilization benefit calculation result is higher than the preset benefit threshold, and the parameter update submodule maintains the median propagation delay. (TTI) and standard deviation of propagation delay (TTI) remains unchanged, and the maximum release ratio of conditionally schedulable resources is slightly increased.

[0093] At the end of period A, the long-term adaptive update module updates its parameters based on the cumulative borrowing revenue records during that period, using an exponentially weighted moving average method. Assuming a smoothing coefficient of 0.3, the average change in latency violation rate after borrowing in this period for the uRLLC slice is... Previous cycle The update magnitude driven by the current cycle borrowing return indicator is: After the update Weighting coefficients of the game utility function Updated from 1.20 to 1.23. The value has been updated from 0.80 to 0.78 and will take effect in B-cycle scheduling.

[0094] Throughout the implementation process, data is generated from event parameters output by the micro-burst event detection module ( , , The interference arrives into the causal propagation analysis module, generating two directed edges for the interference arrival probability distribution. This probability distribution drives the non-uniform elastic reservation module to release conditionally schedulable resources according to time slot differentiation (12 resources are released within the uRLLC slice window, and 7 within the eMBB slice window), and injects them into the shared resource pool. The conditionally schedulable resources cause changes in the resource surplus index of each slice, and the resource index calculation module outputs the resource surplus index of each slice accordingly. and The Nash bargaining game solution module determines the borrowing and lending roles based on these two types of indices and solves the borrowing allocation vector; the security constraint verification module generates physical resource block remapping instructions while ensuring the minimum service threshold of each lender; the conditional resource real-time recovery module completes the time slot boundary return of conditional resources when the interference actually arrives; the final feedback calibration module and the long-term adaptive update module update the propagation delay parameters and weight coefficients at the window level and periodic level, respectively, so that the system continuously improves its prediction of subsequent micro-burst events and borrowing decisions.

[0095] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A 5G-based industrial multimodal network slicing resource allocation system, characterized in that, include: The micro-burst event detection module is configured to perform sliding window change point detection on the resource occupancy sequence of mMTC slices to identify micro-burst events and extract the start time, duration and peak resource occupancy of the micro-burst events. The causal propagation analysis module is configured to use Granger causality tests to analyze the causal propagation path of micro-burst events to uRLLC and eMBB slices, generate a micro-burst causal propagation path diagram, and obtain the interference arrival probability distribution of each time slot within the interference impact prediction time window based on historical propagation delay sample statistics. The non-uniform elastic reservation module is configured to allocate the number of reserved resource blocks according to the ratio of the arrival probability of interference in each time slot to the maximum probability within the window, and mark the difference resource blocks as conditionally schedulable resources and inject them into the shared resource pool. The resource index calculation module is configured to calculate the resource urgency index and resource surplus index of each slice at the time slot level. The Nash bargaining game solution module is configured to solve the resource borrowing allocation vector between slices based on the resource urgency index and the resource surplus index; The conditional resource instant reclamation module is configured to return borrowed conditional schedulable resources to the uRLLC slice when the queue backlog depth of the uRLLC slice exceeds a preset emergency reclamation threshold.

2. The 5G-based industrial multimodal network slicing resource allocation system according to claim 1, characterized in that, The causal propagation analysis module uses the start time of micro-burst events as the starting point. Using the anchor point as a reference, the resource reallocation instruction sequences of uRLLC, eMBB, and mMTC slices within a preset range before and after the anchor point are extracted. Granger causality tests are used to perform hop-by-hop analysis to determine the significance of the causal impact of changes in mMTC slice resource granting on changes in uRLLC and eMBB slice resource granting, as well as the strength and propagation delay values. This generates a micro-burst causal propagation path graph with each slice as a node and directed edges for slice pairs exhibiting significant causal impact. Each directed edge records the causal impact strength and median propagation delay. and the standard deviation of propagation delay .

3. The 5G-based industrial multimodal network slicing resource allocation system according to claim 2, characterized in that, The causal propagation analysis module queries the directed edges in the microburst causal propagation path graph that point to the uRLLC and eMBB slices, respectively, and obtains the corresponding median propagation delay. and the standard deviation of propagation delay The starting boundary of the time window for predicting the impact of interference is determined as follows: The end boundary is ,in The preset coverage factor; for the total coverage within the window For each time slot, based on the historical propagation delay samples accumulated in the micro-burst causal propagation path diagram, the frequency of samples falling into the corresponding delay interval of each time slot is counted. The frequency of each time slot is normalized by dividing by the total number of samples to obtain the interference arrival probability of each time slot. ,in For time slot index, This forms the interference arrival probability distribution; the two directed edges pointing from the mMTC slice to the uRLLC slice and from the mMTC slice to the eMBB slice are each calculated independently for the interference arrival probability distribution.

4. The 5G-based industrial multimodal network slicing resource allocation system according to claim 1, characterized in that, The non-uniform elastic reservation module for the first [unique] within the window For each time slot, allocate the number of reserved physical resource blocks according to the following formula. : ； in The total amount of resource blocks originally reserved. For the first The probability of interference arrival in each time slot, This represents the maximum probability of interference arrival within the window. The total number of time slots within the window; the number of schedulable resource blocks available for release under the conditions of each time slot is... The conditionally schedulable resources are kept with an independent identifier in the shared resource pool to distinguish them from regular schedulable resources.

5. The 5G-based industrial multimodal network slicing resource allocation system according to claim 1, characterized in that, The resource index calculation module calculates the queue backlog depth for each slice. and the number of remaining available resource blocks After Z-score standardization, the resource urgency index is calculated using the following formula. : ； in For the first The latency-sensitive weighting factor corresponding to each slice type, uRLLC slice Higher than eMBB slices and mMTC slices; when The standardized value is zero and When the standardized value is greater than zero, Take it directly as exceeding the emergency threshold The preset upper limit value; The resource index calculation module calculates the resource surplus index using the following formula. : ； in For the first The number of remaining available resource blocks for each slice in the current time slot. To maintain the The minimum number of resource blocks required for the current throughput of a slice. This represents the maximum capacity of resource blocks for this slice type; The calculation incorporates the conditionally schedulable resources into the available resource range of each slice.

6. The 5G-based industrial multimodal network slicing resource allocation system according to claim 5, characterized in that, The Nash bargaining game solving module will use the resource urgency index. Higher than the preset emergency threshold The slice is marked as the resource borrower, and the resource surplus index is used. Exceeding the preset surplus threshold The slice is marked as the resource lender; define the game utility function. : ； in For the borrower The probability of latency violations decreased for each slice after it obtained borrowed resources. For the lender The throughput loss of each slice due to lending out resources To reduce the weighting coefficient for delay violations, This is the throughput loss weighting coefficient; and Before substitution, mean normalization based on the range is performed separately; the Nash bargaining solution is obtained by solving under the constraint of satisfying the range of resource blocks in each slice. The resource borrowing allocation scheme with the maximum value is obtained, where... For the first The game utility function value corresponding to each participating slice.

7. The 5G-based industrial multimodal network slicing resource allocation system according to claim 6, characterized in that, During the borrowing process, the Nash bargaining game solving module distinguishes between regular surplus resources and conditionally schedulable resources based on the resource identifiers in the shared resource pool, and sets the borrowing priority of conditionally schedulable resources to be lower than that of regular surplus resources; when the lender's regular surplus resources are sufficient to meet the borrowing amount specified by the resource borrowing amount allocation vector, only regular surplus resources are borrowed. When regular surplus resources are insufficient to meet the borrowing volume, resources can be scheduled to participate in cross-slice borrowing based on the conditions for activating the difference. The system also includes a security constraint verification module, configured to calculate the number of remaining resource blocks after the borrowing is performed on each lender's slice. ,verify Whether it is valid, among which For the first The number of remaining available resource blocks for each lender slice. For the amount lent out, This represents the minimum service guarantee threshold for this slice type. If the constraint check fails, the loan amount will be reduced to... Any shortfall in lending resulting from the reduction shall be borne by all borrowers in equal proportion to their original allocation.

8. The 5G-based industrial multimodal network slicing resource allocation system according to claim 1, characterized in that, The system also includes a feedback calibration module, configured to, after the interference impact prediction time window ends, calculate the prediction deviation between the actual interference arrival time of each time slot within the window and the predicted peak probability time slot of the interference arrival probability distribution, and use the prediction deviation to update the median propagation delay of the corresponding directed edge in the micro-burst causal propagation path graph. and the standard deviation of propagation delay ; The system calculates the net resource utilization benefit of conditionally borrowed resources by analyzing the borrowing rate, recovery rate, and recovery response latency of conditionally schedulable resources during the window period. The net resource utilization benefit is the percentage of time slots in which conditionally schedulable resources are effectively borrowed and utilized minus the percentage of time slots incurred due to scheduling overhead caused by recovery operations. When the net resource utilization benefit is higher than a preset benefit threshold, the maximum release ratio of conditionally schedulable resources is increased; when the net resource utilization benefit is lower than the preset benefit threshold, the maximum release ratio is reduced.

9. The 5G-based industrial multimodal network slicing resource allocation system according to claim 1, characterized in that, The system also includes a long-term adaptive update module, configured to update the latency-sensitive weighting factor of each slice type at the end of each scheduling cycle based on the accumulated borrowing revenue records during that cycle, using an exponentially weighted moving average method. Weighting coefficients in the game utility function , The borrowing revenue record includes the changes in latency violation rate and throughput of each slice after borrowing, and the updated parameters take effect in subsequent scheduling cycles.

10. A 5G-based industrial multimodal network slice resource allocation method, used to execute the 5G-based industrial multimodal network slice resource allocation system according to any one of claims 1 to 9, characterized in that, include: Step 1: Collect resource grant records and real-time performance indicators of uRLLC slices, eMBB slices and mMTC slices in each transmission time interval through the micro-burst event detection module. Perform sliding window change point detection on the resource occupancy sequence of mMTC slices to identify micro-burst events and extract the start time, duration and peak resource occupancy of micro-burst events. Step 2: Using the causal propagation analysis module, Granger causality test is used to analyze the causal propagation path of micro-burst events to uRLLC slices and eMBB slices, generating a micro-burst causal propagation path diagram, and obtaining the interference arrival probability distribution of each time slot within the interference impact prediction time window based on historical propagation delay sample statistics. Step 3: Allocate the number of reserved resource blocks according to the ratio of the arrival probability of interference in each time slot to the maximum probability within the window through the non-uniform elastic reservation module, and mark the difference resource blocks as conditionally schedulable resources and inject them into the shared resource pool. Step 4: Calculate the resource urgency index and resource surplus index of each slice at the time slot level using the resource index calculation module; Step 5: Solve the resource borrowing allocation vector between slices based on the resource urgency index and resource surplus index using the Nash bargaining game solution module, and perform cross-slice resource borrowing; Step 6: When the queue backlog depth of the uRLLC slice exceeds the preset emergency recovery threshold, the conditional schedulable resources that have been borrowed are returned to the uRLLC slice through the conditional resource real-time recovery module.

Images