Mobile edge node remote state scheduling method, electronic device, and storage medium

Abstract

The application provides a mobile edge node remote state scheduling method, electronic equipment and a storage medium. The method comprises the following steps: acquiring the current information age of each monitored quantity in each time slot; querying a preset index table according to the current information age of each monitored quantity to obtain the scheduling index corresponding to each monitored quantity under the current information age; selecting the monitored quantity with the largest scheduling index as the scheduling object of the current time slot, performing state sampling on the scheduling object, and transmitting the state data obtained by sampling; updating the information age of each monitored quantity according to the sampling result and the transmission result; if the sampling result of the scheduling object is successful and the transmission is successful, resetting the information age of the scheduling object in the next time slot to an initial value; otherwise, increasing the information age of the scheduling object in the next time slot by 1; and increasing the information age of each monitored quantity that is not scheduled in the next time slot by 1. The scheduling method is more scientific, the calculation is more efficient, and the adaptability to the mobile environment is stronger.

Description

Technical Field

[0001] This invention relates to the field of cross-scheduling technology of communication and edge computing, and more specifically, to a remote state scheduling method for mobile edge nodes, an electronic device, and a storage medium. Background Technology

[0002] In mobile edge environments, there are numerous concurrent state monitoring needs among mobile entities, along-line devices, communication links, and edge nodes. Typical monitoring quantities include device operating status, environmental status, and security-related alarms. Traditional communication systems typically use throughput, end-to-end latency, packet loss rate, or queue length as optimization metrics, but these metrics cannot fully and accurately describe the "state freshness" of data in remote monitoring systems. For example, the transmission latency of a single data transmission may be very short, but if a certain monitoring quantity has not been sampled and updated for a long time, the state held by the receiving end will still be outdated. Therefore, Age of Information (AoI) has been introduced to characterize the freshness of the receiver's state information, defined as the difference between the current time and the time when the receiver last successfully received a state packet.

[0003] In typical multi-source state update scenarios, systems are often subject to extremely strict communication resource constraints. Multiple monitoring quantities need to share limited wireless links, and only one monitoring quantity can be scheduled per time slot, leading to fierce resource competition. Furthermore, the service importance of different monitoring quantities varies significantly (for example, alarms related to control loops have a much higher priority than ordinary environmental quantities). This necessitates the introduction of weighted information age for differentiated measurement in system cost assessment.

[0004] Currently, there are two main existing technical solutions to the multi-source state update scheduling problem mentioned above, but both have obvious limitations: Firstly: Fixed polling and periodic update scheme Fixed polling schemes schedule monitoring data in a predetermined, fixed order, while periodic update schemes schedule data according to a predetermined period. The drawback of these schemes is their rigid scheduling strategy, which cannot be adjusted based on real-time information age, business weight, or dynamic link status. This means that even if some high-weight, critical monitoring data has become outdated and faces extremely high monitoring risks within a short period, it must still passively wait for the fixed polling order. Furthermore, the fixed update period is difficult to adapt to the dynamic changes in sampling success rate and wireless environment in actual systems.

[0005] Secondly: Greedy maximum weight AoI scheme

[0006] This scheme calculates the current weighted AoI for all monitored quantities in each time slot and prioritizes scheduling the monitored quantity with the largest current weighted value. The limitation of this scheme lies in its inherently "short-sighted strategy." It only compares the immediate cost of actions and fails to explicitly characterize the long-term impact of the current scheduling action on future AoI evolution through an average cost model. Especially in environments with low effective update success rates, a simple greedy comparison cannot determine whether the current scheduling choice has long-term system benefits. Summary of the Invention

[0007] The present invention aims to provide a method and device for remote state scheduling of mobile edge nodes, so as to solve at least one of the above-mentioned problems.

[0008] In a first aspect, the present invention provides a remote state scheduling method for mobile edge nodes, applicable to scenarios where multiple monitoring quantities share update resources and at most one monitoring quantity is scheduled per time slot, including: In each time slot, the current information age of each monitored quantity is obtained; the information age represents the number of time slots between the current time and the time when the status update of the monitored quantity was last successfully received. Based on the current information age of each monitoring quantity, query the preset index table to obtain the scheduling index corresponding to each monitoring quantity under the current information age; Select the monitoring quantity with the largest scheduling index as the scheduling object of the current time slot, perform state sampling on the scheduling object, and transmit the state data obtained based on the state sampling. If the state sampling is successfully performed and the state data is successfully transmitted, the information age of the next time slot of the scheduled object is reset to the initial value; otherwise, the information age of the next time slot of the scheduled object is incremented by 1, and the information age of the next time slot of each unscheduled monitoring quantity is incremented by 1. The index table is pre-generated as follows: For each monitored quantity, a single-class average cost Markov decision process model is established with information age as the state and scheduling and non-scheduling as actions. In the model, a subsidy parameter is introduced for the non-scheduling action. For each information age state, the subsidy parameter value that makes the long-term cost of scheduling action equal to the long-term cost of non-scheduling action is solved. The subsidy parameter value is used as the scheduling index of the monitored quantity in the corresponding information age state and stored in the index table.

[0009] Preferably, in the remote state update scheduling method, the instantaneous cost of the single-class average cost Markov decision process model adopts a weighted information age, where the weighted information age is the product of the business weight of the monitored quantity and its current information age; different monitored quantities are configured with different business weights, so that monitored quantities with different business weights have different scheduling indices under the same information age.

[0010] Preferably, in the remote state update scheduling method, in the single-class average cost Markov decision process model, the state transition of the scheduling action is based on the probability of successful effective update; The effective update success probability is the product of the sampling success probability and the transmission success probability; After the scheduling action is executed, the probability that the information age is reset to the initial value is the probability of successful effective update, and the probability that the information age is incremented by 1 is the difference between 1 and the probability of successful effective update.

[0011] Preferably, in the remote state update scheduling method, the pre-generation of the index table involves solving the single-class average cost Markov decision process model using a relative value iteration method, including: Initialize the relative value function for each information age state; For each information age state, calculate the non-scheduled action value and the scheduled action value respectively; The non-scheduled action value is the sum of the immediate cost corresponding to the information age state and the relative value of the next state; The scheduling action value is the immediate cost corresponding to the information age state plus a weighted sum of the relative value of the initial state and the relative value of the next state, weighted by the probability of successful effective update. Take the smaller value between the unscheduled action value and the scheduled action value as the new value of the state, and normalize it with the new value of the reference state to obtain the updated relative value function. Determine if the difference between the relative value functions before and after the update is less than the preset convergence threshold; if not, repeat the above calculation; if it is satisfied, output the converged state action values.

[0012] Preferably, in the remote state update scheduling method, for each information age state, a binary search method is used to solve for the subsidy parameter value that makes the long-term cost of scheduling actions equal to that of non-scheduling actions, including: Set the search range for subsidy parameters; Take the midpoint of the search interval as the current subsidy parameter value, solve the single-class average cost Markov decision process model under the subsidy parameter value, and calculate the difference between the scheduling action value and the non-scheduling action value under the information age state. If the difference is negative, the lower bound of the search interval is increased; if the difference is positive, the upper bound of the search interval is decreased. When the length of the search interval or the absolute value of the difference is less than the preset error threshold, the current subsidy parameter value is used as the scheduling index for the age status of the information.

[0013] Preferably, in the remote status update scheduling method, the information age of each monitored quantity is limited to a preset maximum value; When the information age increases to the preset maximum value, it remains at the preset maximum value and does not increase further; when querying the index table, if the current information age of the monitored quantity reaches or exceeds the preset maximum value, the scheduling index corresponding to the preset maximum value is used as the current scheduling index of the monitored quantity.

[0014] Preferably, the remote state update scheduling method further includes: After each scheduling is completed, the sampling success probability estimate of the scheduled object is updated using an exponentially weighted average method based on the sampling results; the transmission success probability estimate of the scheduled object is updated using an exponentially weighted average method based on the transmission results. The current valid update success probability estimate is calculated by multiplying the updated sampling success probability estimate by the transmission success probability estimate. When the deviation between the current effective update success probability estimate and the effective update success probability used when the index table was generated exceeds a preset threshold, the index table is regenerated or switched to an index table corresponding to the current effective update success probability.

[0015] Preferably, in the remote status update scheduling method, during the pre-generation of the index table, the method further includes verifying the indexability of each monitored quantity, including: The optimal strategy of the single-class average cost Markov decision process model is solved under different subsidy parameter values ​​to obtain the set of information age states where the non-scheduled action is optimal under each subsidy parameter value. Check whether the state set monotonically expands as the subsidy parameter value increases; if the monotonically expanding condition is met, then the monitored quantity is scheduled using the exponential table method.

[0016] On the other hand, the present invention also discloses an electronic device, comprising: Memory, used to store computer programs; A processor is configured to execute a computer program stored in the memory to implement the steps of the remote state scheduling method for mobile edge nodes as described in any one of claims 1 to 8.

[0017] On the other hand, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the remote state scheduling method for mobile edge nodes as described in any one of claims 1 to 8.

[0018] As can be seen, this embodiment reflects the long-term scheduling value of each monitored quantity under different information age states through the scheduling index, so that the scheduling decision not only depends on the current instantaneous state, but also implicitly reflects the impact of future information age evolution on system cost, which has a more complete decision basis compared with fixed polling or pure greedy methods.

[0019] Furthermore, in actual implementation, the index table is pre-calculated during the offline phase, and each time slot in the online scheduling phase only needs to perform table lookup and maximum value comparison operations. The computational overhead is linearly related to the number of monitored quantities, which meets the processing latency requirements of mobile edge nodes for real-time decision-making.

[0020] In addition, the information age of the scheduled object is only reset when sampling and transmission are successful; otherwise, it continues to increment, ensuring that the maintenance of the information age is consistent with the actual state reception result and avoiding misjudgment of the information age status in the event of update failure.

[0021] In summary, the scheduling method of this invention is more scientific, more computationally efficient, and more adaptable to mobile environments. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the present invention or related technologies, the accompanying drawings used in the description of the embodiments of the present invention or related technologies will be briefly introduced below. Obviously, the provided drawings are only used to illustrate some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in the provided drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the structure of the update system on which the remote state scheduling method for mobile edge nodes of the present invention is based; Figure 2 This is a flowchart illustrating the steps of an embodiment of the remote state scheduling method for mobile edge nodes according to the present invention. Figure 3 This is an example of an AoI state update activity graph in an embodiment of the mobile edge node remote state scheduling method of the present invention. Figure 4 In this embodiment of the remote state scheduling method for mobile edge nodes of the present invention, an index table generates an activity graph; Figure 5 This is an online scheduling timing diagram in an embodiment of the remote status scheduling method for mobile edge nodes of the present invention; Figure 6 This is a structural block diagram of an embodiment of the electronic device of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] The invention will be described more fully below with reference to the accompanying drawings; however, the embodiments shown may be embodied in different forms, and the invention should not be construed as limited to the embodiments set forth below. Rather, these embodiments are provided so that the invention will be thorough and complete, and that those skilled in the art will fully understand the scope of the invention.

[0026] Reference Figure 1 , Figure 1 The remote status update system for mobile edge nodes, upon which the remote status scheduling method for mobile edge nodes is based, includes multiple monitoring quantities or status sources, a sampling module, a wireless transmission module, a scheduling module, and a monitoring application module. The scheduling module contains an information age status maintenance module, a scheduling decision module, and an index table module. The scheduling module and its related components are deployed in the mobile edge nodes of the system.

[0027] like Figure 1 As shown, the mobile edge node remote status update system is applied to mobile edge monitoring scenarios where multiple monitoring quantities share update resources. These mobile edge monitoring scenarios can include vehicle-to-ground communication monitoring systems, air-to-ground communication monitoring systems, trackside edge computing node monitoring systems, train-side edge node monitoring systems, drone edge node monitoring systems, or industrial IoT remote status monitoring platforms. In other words, mobile edge nodes can be deployed in areas with mobile nodes such as trains, cars, and drones, or as front-end nodes on the trackside, base station side, or central monitoring platform. This invention can be implemented as long as the node can maintain AoI, perform index lookup, and output scheduling objects.

[0028] The mobile edge node remote status update system includes multiple monitoring variables, denoted as monitoring variable 1 to monitoring variable N. Each monitoring variable corresponds to a status source. Multiple monitoring variables share the same sampling resources, wireless transmission resources, or logical scheduling resources, and within each discrete time slot, the system schedules at most one monitoring variable to perform status sampling and status data transmission.

[0029] The mobile edge node implements its functions through the aforementioned scheduling module and its related components. Specifically, the mobile edge node includes an information age status maintenance module, a scheduling decision module, an index table storage module, and an index table generation module. The information age status maintenance module maintains the information age of each monitored quantity in the current time slot; the scheduling decision module queries the index table based on the current information age and selects the scheduling object; the index table storage module stores pre-generated scheduling indices; and the index table generation module generates index tables offline during system initialization or parameter update phases.

[0030] Each monitoring quantity i has a corresponding business weight wi. The business weight wi is used to characterize the importance of the monitoring quantity in the remote monitoring business. For example, security alarm monitoring quantities can be assigned a larger business weight, while general environmental monitoring monitoring quantities can be assigned a smaller business weight. By setting different business weights, different monitoring quantities can correspond to different scheduling indices under the same information age, thereby reflecting differentiated update priorities.

[0031] In this embodiment, the information age Ai(t) represents the number of time slots since the latest state information of the monitored quantity i currently held by the receiving end was generated at time slot t. If the monitored quantity i is scheduled in time slot t and sampling and transmission are successful, the information age of the monitored quantity is reset to the initial value in the next time slot. Preferably, the initial value is 1. If sampling or transmission fails, it is considered that no valid state update has been formed in this time slot, and the information age of the monitored quantity is increased by 1 in the next time slot. The information age of other monitored quantities that have not been scheduled is also increased by 1 in the next time slot.

[0032] To facilitate offline calculations and engineering implementation, this embodiment allows setting a maximum information age. Once the information age of a monitored quantity reaches Amax, it remains at Amax and does not increase further if updates fail. During online table lookups, if the current information age of a monitored quantity is greater than or equal to Amax, the scheduling index corresponding to Amax is used as the current scheduling index for that monitored quantity.

[0033] The functions of each module are as follows: (1) State source Status sources are used to generate status information to be monitored. Each status source corresponds to a monitoring variable, denoted as... The number of monitored items is... Each monitored quantity has a business weight. Business weights are used to represent the importance of monitored quantities in the system's weighted AoI cost.

[0034] (2) Sampling module

[0035] The sampling module performs status sampling after a certain monitored quantity is scheduled. The sampling result has two states: success or failure. The probability of successful sampling is denoted as... When sampling fails, a valid state update packet cannot be generated in this time slot.

[0036] (3) Wireless transmission module

[0037] The wireless transmission module is used to send the status update packet generated after successful sampling to the remote monitoring terminal. The wireless transmission module can employ cellular communication, 5G-R, LTE-R, wireless LAN, or other wireless communication methods. The probability of successful transmission is denoted as... If transmission fails, the update for this time slot will not take effect.

[0038] (4) Scheduling module

[0039] The scheduling module includes: AoI State Maintenance Module The AoI status maintenance module records the age information of each monitored quantity in the current time slot. When monitoring volume In the time slot If scheduling and sampling / transmission are both successful, the AoI for the next time slot is reset to 1. Otherwise, the AoI for the next time slot is incremented by 1. To facilitate numerical solutions, the AoI state can be truncated to between 1 and 1 during the offline computation phase. .

[0040] Index table generation module

[0041] The index table generation module is used to calculate the scheduling index for each monitored quantity under different AoI states offline. This module includes a single-class MDP solution unit, a threshold structure checking unit, a Lagrange subsidy solution unit, and a binary search unit. Preferably, the index table generation module can be set up on a remote computing device, and the generated index table is distributed to the mobile edge node.

[0042] Scheduling decision module

[0043] The scheduling decision module reads the current AoI of all monitored quantities in each time slot, queries the corresponding scheduling index from the index table (e.g., the Whittle index), and selects the monitored quantity with the highest index for scheduling. If multiple monitored quantities have the same index, a fixed number priority, random breaking of ties, or rotation of ties can be adopted.

[0044] (5) Monitoring Application Module

[0045] The monitoring application module uses the latest status information maintained by the mobile edge nodes for display, alarms, control assistance, or upper-layer business processing. This module is not a necessary limiting part of the scheduling method of this invention, but can be part of the system implementation environment.

[0046] Multiple state sources share sampling and wireless transmission resources. The scheduling decision module outputs only one scheduled monitored quantity number per time slot. The sampling module samples only that monitored quantity. If sampling is successful, the wireless transmission module continues to send state packets. If either sampling or transmission fails, the update of that monitored quantity in this time slot fails. The AoI state maintenance module modifies the AoI of all monitored quantities based on the update results. The index table generation / storage module provides the scheduling decision module with the scheduling index corresponding to the current AoI.

[0047] Reference Figure 2 , Figure 2 This is a flowchart illustrating the steps of an embodiment of the remote status scheduling method for mobile edge nodes according to the present invention. It is applied to scenarios where multiple monitoring quantities share update resources and only one monitoring quantity is scheduled per time slot, and is applied to mobile edge nodes. For example, the scheduling module of a mobile edge node includes the following steps: Step 210: In each time slot, obtain the current information age of each monitored quantity; the information age represents the number of time slots between the current time and the time when the status update of the monitored quantity was most recently successfully received.

[0048] Step 220: Based on the current information age of each monitoring quantity, query the preset index table to obtain the scheduling index corresponding to each monitoring quantity under the current information age; Step 230: Select the monitoring quantity with the largest scheduling index as the scheduling object of the current time slot, perform state sampling on the scheduling object, and transmit the sampled state data; Step 240: If the state sampling is successfully performed and the state data is successfully transmitted, the information age of the next time slot of the scheduled object is reset to the initial value; otherwise, the information age of the next time slot of the scheduled object is incremented by 1, and the information age of the next time slot of each unscheduled monitoring quantity is incremented by 1.

[0049] In a discrete time-slot system, at most one monitoring variable can be scheduled per time slot. Let Indicates the amount of monitoring In the time slot Is it scheduled?

[0050] The system resource constraints are:

[0051] This indicates that the system uses available update resources once per time slot by default, therefore a monitoring quantity is selected for scheduling per time slot.

[0052] Monitoring volume Define its time slot The current information age AoI is When the monitoring volume has just been successfully obtained When the latest state is updated, the AoI value in the next time slot is 1. If the update is unsuccessful, the AoI is incremented by 1 in the next time slot.

[0053] Specifically, if the monitoring volume In the time slot If the update is scheduled and successful, then:

[0054] otherwise:

[0055] For other monitoring quantities that are not scheduled, the AoI is also increased:

[0056] Furthermore, the index table is pre-generated in the following manner: for each monitored quantity, a single-class average cost Markov decision process model is established with information age as the state and scheduling and non-scheduling as actions. In the model, a subsidy parameter is introduced for the non-scheduling action. For each information age state, the subsidy parameter value that makes the long-term cost of scheduling action equal to the long-term cost of non-scheduling action is solved. The subsidy parameter value is used as the scheduling index of the monitored quantity in the corresponding information age state and stored in the index table.

[0057] In this embodiment, information age is used to describe the freshness of the status information at the receiving end. The index table is a data structure pre-generated offline before the system runs. It stores the corresponding scheduling index values ​​using the monitoring quantity number and information age status as indexes. During the online scheduling phase of each time slot, the system obtains the scheduling index of each monitoring quantity by reading the current information age of each monitoring quantity and querying the index table, thereby avoiding the repeated execution of complex numerical iterative calculations during the online phase.

[0058] The scheduling index is derived from the solution of a single-class average cost Markov decision process model after introducing a subsidy parameter to the non-scheduled action. Specifically, for a given information age state, as the subsidy parameter increases, the system tends to choose the non-scheduled action; conversely, when the subsidy parameter is small, the system tends to choose the scheduled action. The scheduling index is the subsidy parameter value that makes the long-term cost of the scheduled action exactly equal to the long-term cost of the non-scheduled action. In multi-monitor scheduling scenarios, a larger scheduling index means that, at the current information age state, the demand for scheduling resources for that monitor is relatively higher.

[0059] The single-class average cost Markov decision process model represents the scheduling decision problem of a single monitored quantity as a sequential decision model with information age as the state space and scheduling and non-scheduling as the action space. The model uses long-run average cost as the optimization objective and describes the optimal value of each state through the Bellman equation. By introducing a subsidy parameter for the non-scheduling action, the relative advantage / disadvantage between proactive scheduling and passive non-scheduling changes with the subsidy parameter, thus allowing for a state-by-state solution to obtain the subsidy parameter value that equalizes the long-run costs of both actions.

[0060] The working principle of the scheduling method provided in this embodiment of the invention is as follows: Before system operation, in the index table generation stage, a single-class average cost Markov decision process model is established for each monitored quantity, and the scheduling index is solved state by state to form a two-dimensional index table. During system operation, at the beginning of each time slot, the scheduling decision module reads the current information age of all monitored quantities, queries the corresponding index table, and obtains the current scheduling index of each monitored quantity. The monitored quantity with the largest scheduling index is selected as the scheduling object of the current time slot, and its state sampling and wireless transmission are performed. Based on the sampling results and transmission results, the information age maintenance module updates the information age of both the scheduled and unscheduled monitored quantities.

[0061] As can be seen, this embodiment reflects the long-term scheduling value of each monitored quantity under different information age states through the scheduling index, so that the scheduling decision not only depends on the current instantaneous state, but also implicitly reflects the impact of future information age evolution on system cost, which has a more complete decision basis compared with fixed polling or pure greedy methods.

[0062] Furthermore, in actual implementation, the index table is pre-calculated during the offline phase, and each time slot in the online scheduling phase only needs to perform table lookup and maximum value comparison operations. The computational overhead is linearly related to the number of monitored quantities, which meets the processing latency requirements of mobile edge nodes for real-time decision-making.

[0063] In addition, the information age of the scheduled object is only reset when sampling and transmission are successful; otherwise, it continues to increment, ensuring that the maintenance of the information age is consistent with the actual state reception result and avoiding misjudgment of the information age status in the event of update failure.

[0064] In a preferred embodiment, the instantaneous cost of the single-class average cost Markov decision process model is weighted information age, which is the product of the business weight of the monitored quantity and its current information age; different monitored quantities are configured with different business weights, so that monitored quantities with different business weights have different scheduling indices under the same information age.

[0065] In other words, the immediate cost is defined by weighted information age, specifically as follows:

[0066] in, Indicates the amount of monitoring The business weight is a positive real number parameter configured for each monitored quantity, used to characterize the relative importance of that monitored quantity in the system scheduling objectives. The value of the business weight can be determined based on the security level, service level, or application requirements of the service corresponding to the monitored quantity, and different monitored quantities can be configured with the same or different business weights.

[0067] Indicates the amount of monitoring In the time slot Information age. For time slots The weighted information age, which is the immediate cost.

[0068] The scheduling objective of this embodiment is not to minimize single transmission delay, but to reduce the average weighted information age over long-term operation. The scheduling objective is a long-term average objective, which can be expressed as:

[0069] In practical engineering implementation, the average weighted AoI can be obtained through finite-duration simulation or by using a statistical system with a running window.

[0070] In this embodiment, monitoring quantities with higher business weights have a relatively higher scheduling index when their information age is the same, making them more likely to be selected when competing for scheduling resources with other monitoring quantities. This allows the system to allocate more update opportunities to high-importance monitoring quantities during long-term operation, reducing the long-term average weighted information age of high-weight monitoring quantities. Different monitoring quantities can be flexibly configured with their own business weights, and this embodiment of the invention does not limit the specific numerical range or configuration source of the business weights.

[0071] In a preferred embodiment, the information age of each monitored quantity is limited to a preset maximum value; when the information age increases to the preset maximum value, it remains at the preset maximum value and does not increase further; when querying the index table, if the current information age of the monitored quantity reaches or exceeds the preset maximum value, the scheduling index corresponding to the preset maximum value is used as the current scheduling index of the monitored quantity.

[0072] To prevent the state space from growing indefinitely, a maximum AoI cutoff value can be set. :

[0073] Reset to [value] when the update is successful. .

[0074] The value can be determined based on the maximum allowable state of obsolescence for the monitoring business, computing resources, and simulation accuracy requirements.

[0075] This embodiment enables the offline numerical solution process to be completed within limited computing resources and time by performing finite truncation of the state space, thus meeting the pre-computation feasibility requirements of the engineering system.

[0076] Furthermore, during the online phase, information ages exceeding the truncation value are uniformly mapped to the maximum information age truncation value. This ensures that all monitored quantities can obtain valid scheduling indices from the index table, preventing lookup failures or abnormal scheduling decisions due to information ages exceeding the table's range. The specific value of the maximum information age truncation value can be flexibly set based on the maximum allowable state staleness for the monitoring service and system computing resources; this embodiment of the invention does not impose limitations on this. (Refer to...) Figure 3 , Figure 3 This is a flowchart illustrating the information age (AoI) state update activity of the entire network monitored quantities within a single discrete time slot in an embodiment of the remote state scheduling method for mobile edge nodes of the present invention. The online dynamic evolution of the information age is executed in a closed loop by the information age state maintenance module within the edge node to ensure that the state freshness count held by the receiving end remains consistent with the actual feedback results of physical sampling and channel transmission. In a discrete time slot system, a single time slot... The update process within the organization specifically includes the following steps: In the time slot Initially, the system first performs a current status read operation, which reads or obtains the current information age of all monitored data within the entire network, denoted as the current information age vector. The information age represents the number of time slots between the current moment and the moment when the monitoring status update was most recently successfully received.

[0077] Subsequently, the scheduling decision module performs an online table lookup and matching based on the read current information age and a pre-set offline index table, and executes a maximum value comparison decision to select the monitoring quantity with the largest scheduling index as the scheduling object for the current time slot, denoted as the selected specific monitoring quantity. Once the selection is complete, the system calls its internal hardware interface to send a command to the sampling module targeting the selected object (monitoring quantity). The sampling module only targets the selected monitored quantity, according to the instruction. Perform specific state sampling operations.

[0078] After the sampling operation is completed, the system enters the first-level conditional decision-making node based on the physical layer sampling observation results fed back by the sampling module in a closed loop: 1. If the sampling result of the sampled object is determined to be a sampling failure: This indicates that the monitored quantity... No valid state update packet was generated within this time slot, and the monitored quantity... A remote state update failed. The system control flow skipped the wireless transmission stage and flowed laterally into the information age accumulation and bounded truncation control branch. At this point, the information age state maintenance module performed an information age limit accumulation update for the scheduled object. By incrementing the current information age by 1 and performing a minimum value operation with a preset maximum value, the information age of its next time slot was limited to within the preset maximum value, i.e., it evolved into maintenance. ; wherein, the A preset maximum value (or maximum information age cutoff value) is set to prevent state space divergence during offline numerical solutions and to ensure the feasibility of system pre-calculation.

[0079] 2. If the sampling result of the sampled object is determined to be successful: This indicates that a state data packet containing the state data has been successfully formed. The system control flow moves forward, triggering the wireless transmission module to call the underlying wireless communication link, such as 5G-R, LTE-R, or wireless LAN, to perform wireless transmission operations to send the state data packet; after the wireless transmission is completed, the system further flows into the second-level conditional decision-making node based on the communication layer confirmation information fed back by the remote receiver in a closed loop. In the second-level conditional decision-making node, the system adaptively controls the switching between the following two sub-branches based on the different wireless transmission results: 2. Sub-branch 1 (if transmission successful): This indicates that within a single time slot, the system simultaneously meets the complete triggering conditions for successful state sampling and successful transmission of the state data, and the remote state update is determined to be successfully effective. The system control flow then enters the state reset branch, where the information age state maintenance module resets the information age of the scheduled object in the next time slot to its initial value. In this embodiment of the invention, the initial value is specifically chosen to be 1, i.e., an assignment operation is performed. .

[0080] 3. Sub-branch two (if transmission fails): This indicates that due to deep channel fading, sudden co-channel interference, or multi-user concurrent conflicts, the receiver failed to receive the data and could not obtain a valid state update. Therefore, the remote state update of the scheduled object is still considered a failure. The system control flow will also be incorporated into the aforementioned information age accumulation and bounded truncation control branch. The information age state maintenance module will increment the information age of the scheduled object in the next time slot by 1 and apply an upper limit, i.e., it will also maintain it as... .

[0081] Furthermore, regarding the scheduled object (monitoring quantity) being scheduled... While performing the aforementioned divergent evolution for the state freshness count, in order to ensure the temporal consistency of the monitoring status across the entire network, the information age status maintenance module will simultaneously perform freshness count aging operations on all other monitoring quantities that have not been scheduled within the current time slot. Specifically, for any quantity not equal to... And the un-scheduled monitoring volume that did not receive a service opportunity in this time slot ,in The information age of the corresponding next time slot is uniformly incremented by 1 and protected by an upper limit truncation, meaning that its information age state variable is adaptively and recursively updated. .

[0082] Finally, after the information age of all scheduled and unscheduled monitoring data in the entire network for the next time slot has been fully calculated and written to memory, the current time slot... The process is now complete, and the system state variables seamlessly transition to the next time slot. The mobile edge node will rely on the latest information age status to repeatedly return to the above steps to read new vectors and continue to execute the loop, thereby reducing online overhead and achieving the engineering effect of making the expected average weighted information age of the system stably converge and control it at the lowest level during long-term operation.

[0083] Furthermore, in the single-class average cost Markov decision process model, the state transition of the scheduling action is based on the effective update success probability; the effective update success probability is the product of the sampling success probability and the transmission success probability; after executing the scheduling action, the probability that the information age is reset to the initial value is the effective update success probability, and the probability that the information age is increased by 1 is the difference between 1 and the effective update success probability.

[0084] The probability of a successful update describes the overall probability that a scheduled monitored quantity will actually complete a state update after a scheduling operation. A state update requires the successful completion of two steps: First, the state sampling is successful, and a state update package containing the state data is formed; Second, the wireless channel transmission was successful, and the receiving end confirmed that it had received the latest status information.

[0085] If both steps are successful, the information age is reset to its initial value in the next time slot; if either step fails, the receiver cannot obtain a valid status update, and the information age is incremented by 1 in the next time slot.

[0086] In other words, for monitored quantity i, define the probability of a successful effective update:

[0087] in, For monitoring volume Successful sampling event; For monitoring volume The transmission success event; Indicates whether the monitored quantity i is scheduled in time slot t; For monitoring volume The probability of a successful update that resets the AoI after being scheduled.

[0088] If the sampling success event and the transmission success event are independent, or if the transmission success probability is already expressed as the conditional transmission success probability after sampling success, then it can be written as:

[0089] in, Indicates the amount of monitoring The probability of successful sampling, i.e. ; Indicates the amount of monitoring The probability of successful transmission, i.e. .

[0090] As can be seen, this embodiment combines the sampling success probability and the transmission success probability into an effective update success probability and introduces state transition modeling. The calculation of the scheduling index implicitly reflects the probability that a scheduling operation can bring about an actual information age reset under the current link conditions and sampling conditions, so that the final scheduling decision is consistent with the actual update success conditions.

[0091] Furthermore, in the pre-generation of the index table, the relative value iteration method is used to solve the single-class average cost Markov decision process model, including: Initialize the relative value function for each information age state; For each information age state, calculate the non-scheduled action value and the scheduled action value respectively; The non-scheduled action value is the sum of the immediate cost corresponding to the information age state and the relative value of the next state; The scheduling action value is the immediate cost corresponding to the information age state plus a weighted sum of the relative value of the initial state and the relative value of the next state, weighted by the probability of successful effective update. Take the smaller value between the unscheduled action value and the scheduled action value as the new value of the state, and normalize it with the new value of the reference state to obtain the updated relative value function. Determine if the difference between the relative value functions before and after the update is less than the preset convergence threshold; if not, repeat the above calculation; if it is satisfied, output the converged state action values.

[0092] The following example will provide further explanation.

[0093] To reduce the complexity of solving multiple types of problems, this embodiment first establishes an average cost Markov decision process for a single monitoring quantity. For any monitoring quantity, the information age state is the current AoI:

[0094] Actions include:

[0095] If you choose The monitored quantity will not be updated in this time slot. The next state is:

[0096] If you choose With probability Update successful and in progress. With probability Update failed and entered status:

[0097] in This represents the probability of a successful and effective update of the monitored quantity.

[0098] The Bellman equation for average cost per class is:

[0099] in: Average cost; It is a relative value function; The immediate cost of passive actions; The immediate cost of proactive actions; To effectively update the success rate.

[0100] Without the introduction of passive subsidies, the immediate cost can be taken as:

[0101] That is, regardless of whether this time slot is scheduled, the weighted AoI cost contributed by this monitoring quantity is currently [missing information]. The differences in actions are reflected in the distribution of the next state.

[0102] In one embodiment, the calculation process is as follows: Input: weights Effective success probability Maximum AoI Preset convergence threshold

[0103] Output: Relative value function Optimal action

[0104] Step a1: Initialization , .

[0105] Step a2: Calculate the non-scheduled action value for each state a:

[0106] Step a3: For each state Calculate the scheduling action value:

[0107] Step a4: Take the smaller value as the new value:

[0108] Step a5: Normalize using the reference state:

[0109] Step a6: Determine and Is the difference smaller than .

[0110] If satisfied, output the optimal action; If not satisfied, then let Return to step a2.

[0111] Reference state normalization is used to eliminate constant drift in the relative value function of the average cost MDP. The reference state is optional. Other fixed states can also be selected.

[0112] It is understood that the relative value iteration in this embodiment has numerical convergence in a finite state space, and can solve a single-class average cost Markov decision process within a preset accuracy range, providing a convergent action value basis for subsequent scheduling exponent calculation. The reference state normalization operation ensures that the relative value function does not undergo numerical drift during the iteration process, which helps improve computational stability. The preset convergence threshold can be flexibly set according to the exponent accuracy requirements and computational resources; this embodiment of the invention does not impose any limitations on it.

[0113] Furthermore, for each information age state, a binary search method is used to solve for the subsidy parameter value that makes the long-term costs of scheduled actions equal to those of non-scheduled actions, including: Set the search range for subsidy parameters; Take the midpoint of the search interval as the current subsidy parameter value, solve the single-class average cost Markov decision process model under the subsidy parameter value, and calculate the difference between the scheduling action value and the non-scheduling action value under the information age state. If the difference is negative, the lower bound of the search interval is increased; if the difference is positive, the upper bound of the search interval is decreased. When the length of the search interval or the absolute value of the difference is less than the preset error threshold, the current subsidy parameter value is used as the scheduling index for the age status of the information.

[0114] It is understandable that the binary search method is used to numerically solve for the subsidy parameter value that makes the value of the scheduled action equal to the value of the non-scheduled action. The difference between the value of the scheduled action and the value of the non-scheduled action is a monotonic function of the subsidy parameter: when the subsidy parameter increases, the value of the non-scheduled action decreases because it receives more subsidy, causing the difference to change in the positive direction; when the subsidy parameter decreases, the difference changes in the negative direction. Utilizing this monotonicity, the subsidy parameter value that makes the difference approach zero can be gradually approximated by binary-splitting the search interval of the subsidy parameter.

[0115] Specifically, the working process of this embodiment is as follows: For the information age state of a certain scheduling index to be sought, first set the initial search interval of the subsidy parameter, so that the signs of the difference between the action values ​​at the two ends of the interval are opposite, thereby ensuring that the zero point exists in the interval; Take the midpoint of the search interval as the current subsidy parameter value, solve the single-class average cost Markov decision process model with subsidies under this parameter, and calculate the difference between the scheduling action value and the non-scheduling action value under this information age state. If the difference is negative, it indicates that scheduling actions are better under the current subsidy parameters, and the subsidy needs to be increased to enhance the attractiveness of non-scheduling actions, so the lower bound of the search interval is increased; if the difference is positive, it indicates that non-scheduling actions are better, and the subsidy needs to be reduced, so the upper bound of the search interval is decreased. Repeat the above process until the length of the search interval or the absolute value of the difference is less than the preset error threshold, and store the current midpoint subsidy parameter value as the scheduling index for this state in the index table.

[0116] The following section will further explain the operating mechanism and mathematical principles of the above-mentioned binary search method using a specific slack subproblem model.

[0117] As mentioned earlier, the original multi-class wireless monitoring scheduling problem is constrained by resource coupling, which allows scheduling at most one monitoring quantity per time slot, i.e.:

[0118] This makes it impossible for the scheduling actions of each monitored quantity to be independently selected in an offline distributed manner. In order to decouple these multiple types of constraints, this embodiment adopts the Lagrange relaxation concept to relax the constraint into the objective function, thereby transforming the scheduling decision of each monitored quantity into a single-class relaxation subproblem with a subsidy parameter.

[0119] Specifically, for a single monitored quantity, a non-negative subsidy parameter (or Lagrange multiplier) is introduced. When the system chooses not to schedule the monitored quantity in the current time slot (i.e., performs a passive action). When ), the system will obtain a size of Virtual subsidies, from a cost control perspective, are equivalent to reducing the immediate costs of non-scheduled actions. When choosing to schedule the monitored quantity (i.e., to execute an active action); If the cost is less than 100%, then the subsidy will not be received. Therefore, the immediate cost with subsidy is explicitly defined as:

[0120] Let the subsidy parameter be In the single-class subproblem, the corresponding state-action-value functions are respectively and Then, the Bellman equation for the single-class average cost with subsidies is expressed as:

[0121] in:

[0122] In the formula, The subsidy parameters are Average cost per category at that time; This corresponds to the relative value function; The business weight of the current monitoring volume; This represents the probability of a successful update of the monitored quantity. In some specific communication-constrained embodiments, if it is necessary to further consider the wireless sampling power consumption or signaling overhead of mobile edge nodes, a constant term for scheduling cost can be added to the active action value. At this point, the active action value is expanded to: .

[0123] Based on the above subsidized Bellman equation, for any fixed information age state... Define the long-run cost difference function between actively scheduled actions and passively unscheduled actions. for:

[0124] when When this occurs, it indicates that under the current subsidy level, proactive scheduling has a lower long-term cost, and the optimal decision is scheduling; when This indicates that passive non-scheduling has a lower overall cost after factoring in subsidies, making non-scheduling the optimal decision. The optimal decision boundary point where active scheduling and non-scheduling actions reach a state of "no difference in long-run average cost" satisfies... Critical subsidy parameter value at time That is, it is defined as the state of the monitored quantity. The scheduling index (also known as the Whittle index) is used for scheduling. :

[0125] It is understandable that the above long-run cost difference function It's about subsidy parameters. It is a strictly monotonically increasing function. When the subsidy parameter... When the virtual subsidy increases, the passive action value of the non-scheduled action increases because it receives a larger share of the virtual subsidy. Monotonically decreasing, thereby driving the difference Increase in the positive direction; conversely, when the subsidy parameter... When it decreases, the difference The difference then decreases in the negative direction. Utilizing the absolute monotonicity of the action difference with respect to the subsidy parameter, this embodiment employs a numerical binary search method. By progressively reducing the search interval of the subsidy parameter through binary search, it can efficiently and stably approach the precise zero point that makes the difference approach zero, thereby obtaining the scheduling index.

[0126] In this embodiment, for each information age status Numerical solution for its corresponding scheduling index The specific calculation process is as follows: Input: Information to be solved: age and state Business weight Effective success probability Maximum AoI cutoff value Preset exponent calculation error threshold .

[0127] Output: This state Corresponding scheduling index .

[0128] Step b1: Set the initial search range for subsidy parameters The initial search interval should ensure that it covers the difference function. The sign change. If it satisfies on the boundary Then, the search interval is expanded to both sides by doubling until the signs of the action difference between the two ends of the interval are opposite, so as to ensure that the zero point exists within the search interval.

[0129] Step b2: Calculate the midpoint value of the current search interval, and use it as the virtual subsidy parameter value for the current round:

[0130] Step b3: At the current subsidy parameter value Next, the above-mentioned single-class average cost MDP model with subsidies is solved using the relative value iteration method to obtain the non-scheduled action value in the current state. and scheduling action value .

[0131] Step b4: Calculate the long-run cost difference of the actions at the current midpoint:

[0132] Step b5: Based on the calculated difference The symbol is used to adaptively narrow the search interval: like This indicates the current subsidies. The value is relatively small, and proactive scheduling actions are still superior to non-scheduling actions. Therefore, the subsidy needs to be increased to enhance the attractiveness of non-scheduling actions. Thus, the lower bound of the search interval is raised, making... ; like This indicates the current subsidies. The upper bound of the search range is too large. Under the subsidy incentive, non-scheduled actions are already superior to proactive scheduling, necessitating a reduction in subsidies. Therefore, the upper bound of the search range is lowered, making... ; like If the absolute value of the value approaches 0, then a stop is triggered directly.

[0133] Step b6: Determine the length of the current search interval. Or the absolute value of the difference Is it less than the preset error threshold? If the convergence condition is not met, return to step b2 and repeat the binary search calculation; if the convergence condition is met, stop the iteration and reset the current midpoint subsidy parameter value. As a scheduling index for the age status of this information Output and store in an offline index table.

[0134] Furthermore, the pre-generation process of the index table also includes verifying the indexability of each monitored quantity, including: The optimal strategy of the single-class average cost Markov decision process model is solved under different subsidy parameter values ​​to obtain the set of information age states where the non-scheduled action is optimal under each subsidy parameter value. Check whether the state set monotonically expands as the subsidy parameter value increases; if the monotonically expanding condition is met, then the monitored quantity is scheduled using the exponential table method.

[0135] It is understandable that indexability verification is a necessary preprocessing step to determine whether the scheduling index can be reasonably defined before applying a scheduling index table method (such as Whittle index scheduling) to a certain monitored quantity. This is done by checking the optimal strategy structure of a parameterized single-class problem. Its verification object is the monotonicity of the set of information age states where non-scheduling actions are optimal as a function of subsidy parameter values.

[0136] More specifically, it verifies whether the set of information age states with "non-scheduled action" as the optimal action will exhibit a monotonically expanding inclusion relationship as the subsidy parameter value increases when "non-scheduled action" is given a higher virtual subsidy.

[0137] The following is a further explanation using a specific embodiment.

[0138] For any given subsidy parameter Suppose that, under this subsidy, the optimal strategy obtained by solving the single-class average cost Markov decision process model is to use non-scheduled actions ( The set of information age states that represent the optimal action is: Its mathematical form is expressed as:

[0139] According to the definition of indexability in Whittle's index theory, if for any two different subsidy parameter values... and In order to satisfy In all cases, the following monotonically expanding inclusion relationship of sets is satisfied:

[0140] If the monitored quantity is determined to meet the indexable condition under the current parameter configuration, it can be pre-calculated offline and scheduled online using an exponential table. If the above monotonically expanding inclusion relationship is not met after verification, a preset backup scheduling rule, such as polling or random scheduling, will be used for the monitored quantity.

[0141] In practical implementation, in order to verify the monotonically expanding condition of the above set, this embodiment introduces the "threshold structure" of the optimal strategy for simplified judgment.

[0142] The threshold structure refers to a fixed subsidy parameter. Under these conditions, there exists a state threshold that depends on this subsidy parameter. This makes it possible for any information age state Its optimal decision action exhibits the following monotonically binary distribution: When information age At this time, the optimal action is the non-scheduled action. ; When information age At this time, the optimal action is the proactive scheduling action. .

[0143] Under this strategy with a monotonic threshold structure, the aforementioned optimal set of unscheduled states can be simplified as a finite combination of continuous integers, i.e. At this point, the verification set With subsidy parameters The condition for increasing monotonically can be physically equivalently transformed into: checking the state threshold. Whether it depends on the subsidy parameters The increase is monotonous and does not decrease.

[0144] To extract arbitrary fixed subsidy parameters The threshold structure of the optimal strategy, in this embodiment, adopts the following specific checking and extraction algorithm: Input: The set of states obtained by solving a single subproblem. The corresponding optimal action sequence .

[0145] Output: Whether a threshold structure exists, and the extracted current state threshold. .

[0146] Specifically, the steps include the following: Step c1: According to the information age status from arrive Read the corresponding optimal actions one by one in ascending order. .

[0147] Step c2: During the reading process, find the first action that satisfies the optimal action as the active scheduling action (i.e. ) information age status .

[0148] Step c3: If the optimal action cannot be found after traversing the entire state space... If the state is as shown, it indicates that no active scheduling boundary has been formed under the subsidy parameter. In this case, the state threshold can be set. Denote as the boundary extrapolation value This indicates that the system remains in a passive, non-scheduled state within the current truncation range.

[0149] Step c4: If the first optimal action is found in the state space... If the state is such that, then mark the state as a provisional state threshold. And continue to check for all those that meet the requirements. Are the optimal actions for the subsequent states all... .

[0150] Step c5: If the full-check condition of the subsequent actions in step c4 is completely met, then the current policy is determined to have a monotonic threshold structure, and the extracted state threshold is output. Otherwise, it is determined that the action sequence under the current parameters does not satisfy the monotonic threshold structure.

[0151] Based on the verification that the optimal strategy possesses a threshold structure, this embodiment modifies the subsidy parameters in an offline loop. The threshold sequence is increased incrementally in steps, and steps c1 to c5 above are called to extract the corresponding threshold sequences. As long as the threshold sequence is verified to satisfy (in This allows for a closed-loop proof of the expansion and inclusion relationship of the set, thereby confirming that the system satisfies the indexability condition.

[0152] Preferably, the above-described embodiment of the remote state scheduling method for mobile edge nodes further includes: After each scheduling is completed, the estimated sampling success probability of the scheduled object is updated using an exponentially weighted average method based on the sampling results. The transmission success probability estimate of the scheduled object is updated using an exponentially weighted average method based on the transmission results; the current effective updated success probability estimate is calculated by multiplying the updated sampled success probability estimate by the transmission success probability estimate. When the deviation between the current effective update success probability estimate and the effective update success probability used when the index table was generated exceeds a preset threshold, the index table is regenerated or switched to an index table corresponding to the current effective update success probability.

[0153] The following section will further explain the operating principle of the above-mentioned feedback-based online estimation of probability parameters and dynamic switching mechanism of the exponent table.

[0154] In real-world mobile edge monitoring scenarios, the sampling success rate of various monitored parameters and the wireless channel transmission success rate often experience significant dynamic drift due to continuous location changes of mobile nodes, fluctuations in the remaining energy of edge nodes, sudden co-channel interference, and the time-varying characteristics of sampling sensor hardware. To enable the scheduling strategy to adaptively follow environmental fluctuations, this embodiment deploys an online probability estimation module in the edge nodes to continuously track these physical parameters.

[0155] Specifically, set monitoring volume The estimated probability of successful sampling is The estimated probability of successful wireless transmission is: Its initial default value can be directly adopted from empirical probabilities derived from historical offline statistics. In each discrete time slot... After scheduling is completed, the online estimation module receives the execution results feedback from the sampling module and the wireless transmission module, and uses the exponentially weighted average (EWMA) method to calculate the current scheduled monitored quantity. The probability parameters are recursively updated using a sliding mechanism.

[0156] In the formula, The preset smoothing coefficient is used, and it satisfies... This is used to balance the weighting of long-term historical statistical estimates with the current single observation results; Indicates the current time slot for the monitored quantity The actual observed binary values ​​of the sampling results, if the sampling is successful. If no valid sample is formed (sampling failed), then ; Indicates the current time slot for the monitored quantity The actual observed binary value of the wireless transmission result; if the transmission is successful... If the transmission fails, For each of the other monitored quantities that is not scheduled in the current time slot. (in The corresponding sampling and transmission probability estimates remain unchanged in the current time slot, i.e. , .

[0157] After updating the sampling and transmission probability estimates, the online estimation module recursively calculates the current time slot's scheduled monitoring quantity in real time based on their product. The current effective update success probability estimate :

[0158] To ensure that the offline index table upon which online scheduling decisions rely remains accurately aligned with the current physical channel and device states, the scheduling decision module continuously monitors the deviation between this estimate and the index table's baseline value. Specifically, the system sets a deviation judgment condition:

[0159] In the formula, This indicates the monitoring parameters that were included in the scheduling index table during its generation. The probability of a successful baseline update. This is the preset deviation threshold. When the above absolute value condition is met, it indicates that the current actual drift deviation has exceeded the acceptable error limit due to environmental changes. At this time, the system will automatically trigger a closed-loop response mechanism: either issue a control command to trigger the edge node to rerun the binary search offline to generate an index table for the new probability parameters, or directly read and switch to the pre-stored backup scheduling index table corresponding to the current new probability interval, thereby completing the online adaptive correction of the scheduling strategy.

[0160] The following description further explains the pre-calculation and online table lookup scheduling of the index table and the index table generation process in the embodiments of the present invention.

[0161] To reduce online computation, this embodiment pre-calculates the call index corresponding to the AoI status for each monitored quantity during system initialization or parameter update phases; that is, the Whittle index.

[0162] in This indicates the monitored quantity number. The index table can be stored as a two-dimensional table:

[0163] The online scheduling process is as follows: Input: Current AoI vector Index table

[0164] Output: Monitoring data for this time slot

[0165] Step d1: Read the current AoI for each monitored quantity .

[0166] Step d2: If The state used for table lookup is then retrieved. Otherwise take .

[0167]

[0168] Step d3: Query the current index for each monitored quantity:

[0169] Step d4: Select the monitoring quantity with the largest index:

[0170] Step d5: Output Provide the sampling module and the wireless transmission module.

[0171] Step d6: Update all monitored quantities (AoI) based on the sampling and transmission results.

[0172] The Bellman equation is no longer solved during the online phase, nor is a binary search performed. If the number of monitored quantities is... Online scheduling is mainly for The time complexity is approximately [number] table lookups and comparisons. .

[0173] Reference Figure 4 , Figure 4 This is an activity flowchart illustrating the generation of the scheduling index table (i.e., the Whittle index table) during the offline phase in an embodiment of the remote state scheduling method for mobile edge nodes of the present invention. The scheduling index table is pre-generated offline by the index table generation module, which eliminates the overhead of real-time joint dynamic programming in the online scheduling phase. The specific calculation and workflow includes the following steps: First, the system performs a parameter input operation, inputting preset offline system parameters, which include at least the total number of monitored quantities within the system. Various monitoring quantities Business weight Probability of successful baseline update and the maximum information age cutoff value .

[0174] Subsequently, the system enters the global loop processing flow, first performing the monitoring quantity selection operation, selecting the specific monitoring quantity in the current scheduling index table to be calculated. ,in .

[0175] For the selected monitoring quantity The system performs a model building operation, independently establishing a single-class average cost Markov decision process model for it. The model uses long-run average cost as its optimization objective, and its information age state space is truncated within a finite range, that is, the state is defined as the current information age. Its action space specifically includes passive, non-dispatched actions (denoted as...). ) and active scheduling actions (denoted as ).

[0176] Next, the system performs numerical solution operations for the subsidized model. Specifically, using the Lagrange relaxation approach, a subsidy parameter is introduced into the passive non-scheduled actions in the model. By changing the subsidy parameters in an offline loop. The values ​​of are used to numerically iterate and solve the single-class average cost Markov decision process model with subsidies by calling the relative value iteration method, thereby calculating and obtaining the values ​​of under specific subsidy parameters. Below, this monitoring quantity corresponds to the age status of each piece of information. scheduling action value and non-scheduled action values .

[0177] Based on the obtained state action values, for each independent information age state Perform numerical binary search operations separately. The binary search aims to find the critical indifference subsidy point where the long-run cost of the scheduled action is exactly equal to the long-run cost of the non-scheduled action, i.e., to solve for the subsidy parameter. nonlinear equations During the binary search process, the system adaptively raises the lower bound of the search interval or lowers the upper bound of the search interval based on the monotonicity of the sign of the difference in action values ​​at the current midpoint, until the length of the search interval or the absolute value of the difference is less than a preset error threshold. .

[0178] When the binary search meets the convergence condition, the system performs an exponential extraction operation, using the currently converged midpoint subsidy parameter value as the monitoring quantity. Age status in the corresponding information The scheduling index below .

[0179] Subsequently, the system performs a single-table save operation, storing the obtained scheduling index sequence in a structured concatenated manner according to the one-to-one correspondence between information age and status, and saving it as the monitoring quantity. A dedicated offline scheduling index table.

[0180] After saving the index table of the current monitored volume, the system proceeds to the conditional branch decision node to check if there are still any unprocessed monitored volumes across the entire network: If the determination result indicates that there are still unprocessed monitored items: the system then proceeds to the "Unprocessed Monitored Items" branch on the left, executes the operation to process the next monitored item, increments the monitored item index number, and returns to "Select Monitored Item". The process of "" is repeated for the next state source, following the steps above. If the determination result is that all monitoring data has been processed, the system will then proceed to the "All monitoring data processed" branch on the right, end the offline calculation loop, and output a complete two-dimensional scheduling index table (i.e., output the complete joint lookup table data structure index_table for all monitoring data in all bounded information age states).

[0181] After the offline calculation process is completed, the complete two-dimensional index table is distributed and statically stored in the index table module of the mobile edge node, so that the online scheduling decision module can perform high-speed table lookup and maximum value comparison operations at the beginning of each discrete time slot.

[0182] Reference Figure 5 , Figure 5 This is a timing diagram illustrating the online scheduling operation in an embodiment of the mobile edge node remote state scheduling method of the present invention. In a discrete time-slot system, in a single time slot... Within this framework, the remote state scheduling method for mobile edge nodes achieves low-complexity online table lookup scheduling through dynamic interaction and state evolution between various functional components. Its specific execution flow includes the following steps: Step 1: In the time slot Initially, the AoI state maintenance module outputs the current information age (i.e., the current information age vector) of all monitored quantities within the current time slot system to the scheduling decision module. , This allows us to obtain the current information age of each monitored quantity. The information age represents the number of time slots between the current moment and the moment when the status update of the monitored quantity was most recently successfully received.

[0183] Step 2: After receiving the current information age vector, the scheduling decision module assigns it a monitoring quantity number. and current information age As an index, the system queries the pre-set index table, specifically the two-dimensional index table `index_table` stored in the index table module, to obtain the scheduling index corresponding to each monitored quantity under the current information age. During this process, if the current information age of a certain monitored quantity reaches or exceeds the preset maximum value... Then, when looking up the table, the truncation mapping is automatically performed (i.e., ), and with the preset maximum value The corresponding scheduling index is used as the current scheduling index for the monitored quantity to ensure that the state space is bounded and the table lookup process is safe and stable.

[0184] Step 3: After obtaining the scheduling indices of all monitored quantities, the scheduling decision module internally performs a maximum value comparison operation, selecting the monitored quantity with the largest scheduling index as the scheduling object for the current time slot.

[0185] Locking specific monitoring volume Subsequently, the scheduling decision module outputs a message to the sampling module containing the scheduling object (monitoring quantity). () Dispatch instructions with numbers.

[0186] Step 4: After receiving the scheduling instruction, the sampling module only targets the selected scheduling object (monitoring quantity). Perform state sampling.

[0187] Step 5: After sampling is completed, the sampling module feeds back the sampling result of this sampling to the scheduling decision module and subsequent modules, that is, returns a binary signal indicating whether the sampling was successful or failed.

[0188] Subsequently, based on the sampling results returned by the sampling module in step 5, the system enters different execution condition branches: If the sampling result of the sampled object is successful, the sampling module will encapsulate the corresponding state data obtained based on the state sampling into a state packet and transmit it to the wireless transmission module. The wireless transmission module executes step 6, calling the wireless communication link to send the state packet to the remote monitoring end; and after the transmission is completed, it executes step 7, feeding back the wireless transmission result to the system closed loop, that is, returning a signal indicating successful or failed transmission. In this embodiment of the invention, whether the scheduled object can ultimately achieve an effective reset of its state freshness (information age) at the receiving end after being scheduled depends on the product of the sampling success probability and the transmission success probability, that is, the effective update success probability.

[0189] If the sampling result of the sampling object is a failure, it is determined that a valid state update cannot be formed in this time slot. The system does not trigger the wireless transmission module to perform the wireless transmission action in step 6, and directly determines that the remote state update of the scheduling object has failed.

[0190] Before the end of the time slot, the sampling results and transmission results are comprehensively and loop-wise fed back to the AoI status maintenance module. The AoI status maintenance module synchronously updates the information age of all monitored quantities across the entire network based on the actual update results, specifically including: If the sampling result of the scheduled object is successful and the transmission is successful (i.e., the state sampling was successfully performed and the state data was successfully transmitted): then the AoI state maintenance module executes step 8, resetting the information age of the scheduled object in the next time slot to the initial value (in this embodiment of the invention, the initial value is specifically maintained as 1). Right now ).

[0191] If the sampling result of the scheduled object is a failure or transmission failure (i.e., any link fails or a failure exists): then it is determined that this state update is ineffective, and the AoI state maintenance module also executes step 8, incrementing the information age of the scheduled object in the next time slot by 1, and introducing a preset maximum value upper limit truncation protection, that is, maintaining it as... .

[0192] While updating the scheduling object: the AoI state maintenance module executes step 10, incrementing the information age of each of the other unscheduled monitoring quantities in the current time slot by 1 for the next time slot. Monitoring volume Its information age evolution is This ensures the consistency of the time sequence for counting the freshness of the entire network status.

[0193] Once the age updates for all monitored data across the entire network are completely written to storage by the AoI status maintenance module, the current time slot... The system state variables have seamlessly transitioned and entered the next time slot. .

[0194] The mobile edge node will repeatedly execute steps 1 to 10 above, so that each time slot only needs to execute [step 1] without performing online global dynamic programming. The table lookup and comparison with the maximum value ensure that the online scheduling overhead is linearly related to the number of monitored items (i.e., the complexity is approximately...). This meets the engineering deployment requirements of mobile edge nodes for real-time status updates and extremely low processing latency.

[0195] In summary: Compared to fixed polling and periodic updates, the scheduling action of this invention is jointly determined by the current AoI, service weight, and the probability of successful effective updates. Fixed rules only guarantee the formal allocation of service opportunities and cannot adjust the update targets based on the aging of the state. This invention reads the current AoI of each monitored quantity in each time slot and reflects the long-term value of proactive updates under the current state through an index table. Therefore, it can give higher scheduling preference to monitored quantities with high weight and older states.

[0196] The greatest of all is simple greed. Compared to other solutions, this invention does not only compare immediate costs. The single-class MDP incorporates the probability of a successful reset after active scheduling, the increase in AoI after failure, and the value of future states into the action value calculation. The Whittle index comes from the indifferent subsidy for active and passive actions in the long-term average cost sense. This mechanism can link the current state with the future evolution of AoI. The online phase still maintains the lookup and comparison format, eliminating the need for global dynamic programming solutions.

[0197] Compared to directly solving a multi-class joint MDP, this invention decomposes the coupled problem into multiple single-class subproblems through Lagrange relaxation and places the main computation in the offline exponent table generation stage. Online scheduling only requires traversing N monitored variables to perform table lookups and maximum value selection. This structure is more suitable for mobile edge nodes to execute within discrete time slots.

[0198] Under the same monitoring quantity, sampling success probability, and transmission success probability, this invention can be compared with fixed polling, periodic updating, random scheduling, and maximum weighted AoI greedy schemes through simulation. Comparison metrics include long-term average weighted AoI, average AoI of high-weight monitoring quantities, peak AoI, online decision-making time, and exponential table storage overhead. Since only table lookup and maximum value comparison are performed in the online phase, the online complexity of this invention is O(n log n). It is suitable for real-time execution on mobile edge nodes.

[0199] The present invention also provides an electronic device, such as Figure 6 As shown, the electronic device includes a processor 111 and a memory 222. The memory 222 stores a computer program, which is executed by the processor 111 to cause the processor 111 to execute the remote state scheduling method for mobile edge nodes provided by the present invention.

[0200] It should be understood that the electronic device provided by the present invention may also include other structures, and the electronic device may also include corresponding modules / units, components / parts and other structures according to its functions. Those skilled in the art can set other structures of the electronic device according to needs or actual application scenarios. The present invention does not specifically limit other structures of the electronic device.

[0201] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to cause the processor to perform the remote state scheduling method for mobile edge nodes provided by the present invention.

[0202] It should be understood that in the remote state scheduling method for mobile edge nodes provided by the present invention, the execution order of each step is not limited to the order described above in conjunction with the accompanying drawings. Without departing from the scope of the present invention, the execution order of each step can be adjusted, and some steps can be executed in parallel.

[0203] Furthermore, those skilled in the art will understand that all or part of the steps / operations in the methods described in this invention, and all or part of the elements / components in the devices, can be implemented as software, firmware, hardware, and suitable combinations thereof. In hardware implementations, the division between functions mentioned in the above description does not necessarily correspond to the division of physical components. For example, a physical component may have multiple functions, or a function may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor (such as a central processing unit, digital signal processor, or microprocessor), or as hardware, or as integrated circuits, such as application-specific integrated circuits (ASICs). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0204] This invention has disclosed exemplary embodiments, and although specific terminology has been used, it is for illustrative purposes only and should be interpreted in a general sense only, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A remote state scheduling method for mobile edge nodes, applied to scenarios where multiple monitoring quantities share update resources and only one monitoring quantity is scheduled per time slot, characterized in that, include: In each time slot, obtain the current age information of each monitored quantity; The information age represents the number of time slots between the current moment and the moment when the monitoring status update was last successfully received. Based on the current information age of each monitoring quantity, query the preset index table to obtain the scheduling index corresponding to each monitoring quantity under the current information age; Select the monitoring quantity with the largest scheduling index as the scheduling object of the current time slot, perform state sampling on the scheduling object, and transmit the state data obtained based on the state sampling. If state sampling is successfully performed and the state data is successfully transmitted, the information age of the next time slot of the scheduling object is reset to the initial value; Otherwise, increment the information age of the scheduled object in the next time slot by 1, and increment the information age of each unscheduled monitoring quantity in the next time slot by 1; The index table is pre-generated as follows: For each monitored quantity, a single-class average cost Markov decision process model is established with information age as the state and scheduling and non-scheduling as actions. In the model, a subsidy parameter is introduced for the non-scheduling action. For each information age state, the subsidy parameter value that makes the long-term cost of scheduling action equal to the long-term cost of non-scheduling action is solved. The subsidy parameter value is used as the scheduling index of the monitored quantity in the corresponding information age state and stored in the index table.

2. The remote state scheduling method according to claim 1, characterized in that, The instantaneous cost of the single-class average cost Markov decision process model adopts a weighted information age, which is the product of the business weight of the monitored quantity and its current information age. Different monitoring quantities are configured with different business weights, so that monitoring quantities with different business weights have different scheduling indices under the same information age.

3. The remote state scheduling method according to claim 1, characterized in that, In the single-class average cost Markov decision process model, the state transition of scheduling actions is based on the probability of successful effective update. The effective update success probability is the product of the sampling success probability and the transmission success probability; After the scheduling action is executed, the probability that the information age is reset to the initial value is the probability of successful effective update, and the probability that the information age is incremented by 1 is the difference between 1 and the probability of successful effective update.

4. The remote state scheduling method according to claim 3, characterized in that, In the pre-generation of the index table, the relative value iteration method is used to solve the single-class average cost Markov decision process model, including: Initialize the relative value function for each information age state; For each information age state, calculate the non-scheduled action value and the scheduled action value respectively; The non-scheduled action value is the sum of the immediate cost corresponding to the information age state and the relative value of the next state; The scheduling action value is the immediate cost corresponding to the information age state plus a weighted sum of the relative value of the initial state and the relative value of the next state, weighted by the probability of successful effective update. Take the smaller value between the unscheduled action value and the scheduled action value as the new value of the state, and normalize it with the new value of the reference state to obtain the updated relative value function. Determine if the difference between the relative value functions before and after the update is less than the preset convergence threshold; if not, repeat the above calculation; if it is satisfied, output the converged state action values.

5. The remote state scheduling method according to claim 4, characterized in that, For each information age state, a binary search method is used to find the subsidy parameter value that makes the long-term costs of scheduled actions equal to those of non-scheduled actions. This includes: Set the search range for subsidy parameters; Take the midpoint of the search interval as the current subsidy parameter value, solve the single-class average cost Markov decision process model under the subsidy parameter value, and calculate the difference between the scheduling action value and the non-scheduling action value under the information age state. If the difference is negative, the lower bound of the search interval is increased; if the difference is positive, the upper bound of the search interval is decreased. When the length of the search interval or the absolute value of the difference is less than the preset error threshold, the current subsidy parameter value is used as the scheduling index for the age status of the information.

6. The remote state scheduling method according to claim 1, characterized in that, The age limit for each monitored quantity is limited to a preset maximum value; When the information age increases to the preset maximum value, it remains at the preset maximum value and no longer increases; When querying the index table, if the current information age of the monitored quantity reaches or exceeds the preset maximum value, the scheduling index corresponding to the preset maximum value is used as the current scheduling index of the monitored quantity.

7. The remote state scheduling method according to claim 3, characterized in that, Also includes: After each scheduling is completed, the sampling success probability estimate of the scheduled object is updated based on an exponentially weighted average method; The estimated probability of successful transmission of the scheduled object is updated based on an exponentially weighted average method. The current valid update success probability estimate is calculated by multiplying the updated sampling success probability estimate by the transmission success probability estimate. When the deviation between the current effective update success probability estimate and the effective update success probability used when the index table was generated exceeds a preset threshold, the index table is regenerated or switched to an index table corresponding to the current effective update success probability.

8. The remote state scheduling method according to claim 1, characterized in that, The process of pre-generating the index table also includes verifying the indexability of each monitored quantity, including: The optimal strategy of the single-class average cost Markov decision process model is solved under different subsidy parameter values ​​to obtain the set of information age states where the non-scheduled action is optimal under each subsidy parameter value. Check whether the state set monotonically expands as the subsidy parameter value increases; if the monotonically expanding condition is met, then the monitored quantity is scheduled using the exponential table method.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute a computer program stored in the memory to implement the steps of the remote state scheduling method for mobile edge nodes as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the remote state scheduling method for mobile edge nodes as described in any one of claims 1 to 8.

Images