Pon-mec service migration method and apparatus, electronic device, and storage medium

Abstract

The application provides a PON-MEC service migration method and device, electronic equipment and storage medium, relates to the technical field of optical communication and mobile edge computing, and the method comprises the following steps: according to a sparse network state observation model, sampling and processing multi-dimensional context attributes collected from a PON-MEC system to form sparse network state observation information; based on tensor column decomposition, inferring and completing missing data in the sparse network state observation information to reconstruct a complete global network state; according to the reconstructed complete global network state, evaluating the comprehensive service quality cost paid for service migration between different mobile edge computing nodes; taking minimizing long-term comprehensive service quality cost as the target, generating an optimal service migration strategy by using a deep Q network model, and scheduling the corresponding mobile edge computing node to perform service migration according to the optimal service migration strategy. The method provided by the application guarantees the business continuity and deterministic transmission demand when a mobile terminal switches across regions.

Description

Technical Field

[0001] This invention relates to the fields of optical communication and mobile edge computing technology, and in particular to a PON-MEC service migration method, apparatus, electronic device and storage medium. Background Technology

[0002] With the rapid development of Industry 4.0 and the Industrial Internet of Things (IIoT), mobile devices such as Automated Guided Vehicles (AGVs) and collaborative robots are widely used in smart manufacturing, placing extremely stringent requirements on network transmission and service computing for low latency and low jitter. The Passive Optical Network (PON) and Mobile Edge Computing (MEC) converged architecture (PON-MEC) effectively reduces end-to-end latency and alleviates core network bandwidth pressure by offloading computing resources at the Optical Network Unit (ONU) level, becoming an ideal solution for supporting high-frequency mobile industrial applications. When terminal devices move across regions, the system needs to dynamically migrate computing tasks between different mobile edge computing nodes to ensure service continuity. However, in harsh industrial environments, frequent status polling and telemetry consume significant uplink bandwidth and introduce substantial signaling overhead. Furthermore, physical factors such as electromagnetic interference and link disruptions in industrial settings often lead to the loss of telemetry data packets, resulting in sparse and incomplete global network status data acquired by the centralized controller.

[0003] Existing service migration scheduling strategies are generally based on the ideal assumption of complete global state visibility. When faced with missing network state data, these methods cannot accurately assess the current system environment and potential service quality costs, leading to serious biases in migration decisions. This can result in high latency, high jitter, or even service interruptions, failing to meet the stringent reliability and determinism requirements of time-sensitive industrial applications. Therefore, a solution is urgently needed to address this problem. Summary of the Invention

[0004] This invention provides a PON-MEC service migration method, apparatus, electronic device, and storage medium to address the deficiencies in the prior art.

[0005] This invention provides a PON-MEC service migration method, applied to industrial passive optical networks and mobile edge computing PON-MEC systems where network status data is missing. The method includes: According to the preset sparse network state observation model, the multi-dimensional context attributes collected from the PON-MEC system are sampled and processed to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information are deduced and completed to reconstruct the complete global network state. Based on the reconstructed complete global network state, assess the overall quality of service cost required for migrating services between different mobile edge computing nodes; With the goal of minimizing the long-term overall service quality cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy.

[0006] According to a PON-MEC service migration method provided by the present invention, the multi-dimensional context attributes include at least one of service type, optical network unit coverage area, user mobility mode, and central processing unit load status of mobile edge computing node; the preset sampling location set is predetermined based on network telemetry capability and signaling overhead constraints.

[0007] According to a PON-MEC service migration method provided by the present invention, the step of reconstructing the complete global network state by extrapolating and completing the missing data in the sparse network state observation information based on tensor column decomposition includes: The high-dimensional tensor of the complete global network state to be reconstructed is decomposed into a product of multiple third-order core tensors. An optimization problem is constructed with the objective of minimizing the error between the sparse network state observation information and the reconstructed data at the preset sampling location set; By iteratively updating the multiple third-order core tensors, the optimization problem is solved to deduce the missing data at unsampled locations and obtain the complete global network state.

[0008] According to a PON-MEC service migration method provided by the present invention, the overall quality of service cost is determined by a weighted sum of service latency and energy consumption; wherein, the service latency includes wireless transmission latency, computation and processing latency of the target mobile edge computing node, and fiber optic backhaul latency of the passive optical network system.

[0009] According to the PON-MEC service migration method provided by the present invention, the deep Q network model includes an evaluation network and a target network; The method of generating an optimal service migration strategy using a deep Q-network model with the objective of minimizing the long-term overall service quality cost includes: The complete global network state is input into the evaluation network, and the action value corresponding to each candidate migration action is output. By interacting with the network environment, state transition samples containing the current state, the migration action performed, the immediate reward obtained, and the next state are collected, and the state transition samples are stored in the experience replay pool; wherein, the immediate reward is the negative value of the overall service quality cost; The parameters of the evaluation network are updated by extracting target state transition samples from the experience replay pool and minimizing the difference between the action value output by the evaluation network and the target action value calculated based on the target network. Based on the updated evaluation network, the migration action that minimizes the long-term expected overall service quality cost is determined, and the optimal service migration strategy is generated.

[0010] According to a PON-MEC service migration method provided by the present invention, after scheduling the corresponding mobile edge computing node to perform service migration according to the optimal service migration strategy, the method further includes: The sparse network state observation information and the deep Q-network model are updated according to the dynamic changes in the network environment.

[0011] According to a PON-MEC service migration method provided by the present invention, the step of updating the sparse network state observation information and the deep Q-network model according to dynamic changes in the network environment includes: New multidimensional context attributes are acquired according to a preset period, and updated sparse network state observation information is formed based on the sparse network state observation model. Based on the updated sparse network state observation information, the tensor column decomposition is performed again to infer and complete the missing data, and the updated complete global network state is obtained. Based on the updated complete global network state, the deep Q-network model is incrementally trained to dynamically adapt to changes in the network environment.

[0012] This invention also provides a PON-MEC service migration device, applied to an industrial passive optical network and mobile edge computing PON-MEC system where network status data is missing. The device includes: The sampling module is used to sample and process the multi-dimensional context attributes collected from the PON-MEC system according to the preset sparse network state observation model to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. The reconstruction module is used to infer and complete the missing data in the sparse network state observation information based on tensor column decomposition, and reconstruct the complete global network state. The evaluation module is used to evaluate the overall quality of service cost required for migrating services between different mobile edge computing nodes based on the reconstructed complete global network state. The migration module is used to generate an optimal service migration strategy using a deep Q-network model with the goal of minimizing the long-term overall service quality cost, and to schedule the corresponding mobile edge computing nodes to perform service migration according to the optimal service migration strategy.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described PON-MEC service migration methods.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the PON-MEC service migration method as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described PON-MEC service migration methods.

[0016] This invention provides a PON-MEC service migration method, apparatus, electronic device, and storage medium, applied to industrial passive optical networks and mobile edge computing PON-MEC systems with missing network state data. It samples multi-dimensional context attributes collected from the PON-MEC system according to a preset sparse network state observation model to form sparse network state observation information. The sparse network state observation model retains data at observable locations through a preset set of sampling locations, while unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information is deduced and completed to reconstruct a complete global network state. Based on the reconstructed complete global network state, the overall quality of service cost required for service migration between different mobile edge computing nodes is evaluated. With the goal of minimizing the long-term overall quality of service cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy. Therefore, this invention first obtains missing observation information based on a pre-defined sparse network state observation model, and then uses tensor column decomposition to extrapolate and complete the missing data, reconstructing a complete global network state, thereby eliminating blind spots in state awareness. Based on this, it evaluates the overall quality of service cost of service migration and generates an optimal migration strategy using a deep Q-network with the goal of minimizing the long-term overall cost. The synergy of these technical means enables the system to accurately perceive the overall network load and resource distribution even under harsh conditions such as limited telemetry and incomplete data in industrial settings, making forward-looking and stable migration decisions, significantly reducing average service latency and transmission jitter, and ensuring service continuity and deterministic transmission requirements when mobile terminals switch regions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the PON-MEC service migration method provided by the present invention.

[0019] Figure 2 This is a schematic diagram of the industrial network system architecture based on PON-MEC and the execution framework of the TT-DQN algorithm provided by this invention.

[0020] Figure 3 This is a schematic diagram comparing the average latency performance under different network load conditions provided by the present invention.

[0021] Figure 4 This is a schematic diagram comparing latency jitter performance under different network load conditions provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the structure of the PON-MEC service migration device provided by the present invention.

[0023] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] The following description, in conjunction with the accompanying drawings, describes a PON-MEC service migration method, apparatus, electronic device, and storage medium of the present invention.

[0026] Figure 1 This is a flowchart illustrating the PON-MEC service migration method provided by the present invention, as follows: Figure 1 As shown, this method, applied to industrial passive optical networks and mobile edge computing (PON-MEC) systems where network status data is missing, includes the following: Step 100: Based on the preset sparse network state observation model, the multidimensional context attributes collected from the PON-MEC system are sampled and processed to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data.

[0027] In this embodiment of the invention, the execution subject of the method is a centralized controller deployed in an optical line terminal or industrial network management system. The centralized controller is responsible for uniformly collecting network status information, performing tensor column decomposition and completion operations, running deep Q network model inference and training, and issuing service migration scheduling instructions to each mobile edge computing node.

[0028] It should be noted that the multidimensional context attributes include information such as the service type of the terminal device, the network topology of the optical network unit (ONU) coverage area, the user's mobility mode, and the central processing unit load status and available computing resource capacity of the mobile edge computing (MEC) node; the preset sampling location set is predetermined based on network telemetry capabilities and signaling overhead constraints.

[0029] Specifically, the sparse network state observation model essentially defines a preset set of sampling locations. Due to severe electromagnetic interference, intermittent link interruptions, and discontinuous telemetry mechanisms adopted to reduce signaling overhead in industrial sites, the system can only successfully retain valid state data at the aforementioned preset observable locations. Data is missing at other locations that are not covered by sampling or fail to be sampled, ultimately forming sparse network state observation information containing a large number of unknown elements.

[0030] Define the complete global network state as a high-dimensional tensor. Due to limitations imposed by industrial site interference and signaling overhead, the actual sparse observation tensors acquired are limited. satisfy: (1) in, For sampling operators, only at the set of observable locations The original data is retained in one location, while data is missing in other locations.

[0031] Step 200: Based on tensor column decomposition, infer and complete the missing data in the sparse network state observation information to reconstruct the complete global network state.

[0032] Specifically, the system uses tensor sequence (TT) decomposition technology to infer and complete the missing parts of the above sparse observation information. Its core idea is to approximate a high-dimensional complete global network state tensor as a sequence multiplication of multiple small-scale third-order core tensors. With the optimization objective of minimizing the deviation between the reconstructed data and the original observation data at known observable locations, the parameters of each core tensor are updated alternately and iteratively to gradually infer the network state values ​​hidden at the originally unsampled locations, thereby reconstructing a smooth, complete, and truly reflective global network state that reflects the load distribution and resource status of the entire network.

[0033] High-dimensional state tensor It is decomposed into the product of multiple third-order core tensors, and the missing data is deduced by minimizing the reconstruction error. The optimization objective function is expressed as: (2) The core tensor is iteratively updated using alternating least squares or gradient descent algorithms, and the reconstructed complete global network state tensor is output.

[0034] Step 300: Based on the reconstructed complete global network state, assess the overall quality of service cost required for migrating services between different mobile edge computing nodes.

[0035] It should be noted that the overall quality of service cost is determined by a weighted sum of service latency and energy consumption; wherein, the service latency includes the wireless transmission latency between the mobile device and the wireless access point (AP), the task queuing and computing processing latency of the target mobile edge computing node, and the fiber optic backhaul latency of the passive optical network system.

[0036] Specifically, the system assesses the overall quality of service cost required to migrate a computing service from its current mobile edge computing node to other candidate target nodes. This overall cost is typically derived by weighted summation of two key indicators: service latency and energy consumption. Service latency further includes the air interface transmission latency between the mobile terminal and the wireless access point, the time overhead of task queuing and central processing unit computation within the target edge computing node, and the fiber optic propagation latency caused by data backhaul through passive optical network fiber optic links between different optical network units. This comprehensively characterizes the potential impact of the migration action on user experience and system efficiency.

[0037] Regarding the first A migration task, which is located at the target node. Overall QoS cost From total delay With total energy consumption The weighted calculation yields: (3) in, and This is a normalization parameter used to adjust the weighting of latency and energy consumption. Latency It covers wireless transmission latency, edge computing node computing processing latency, and fiber optic backhaul latency of passive optical network systems.

[0038] Step 400: With the goal of minimizing the long-term overall service quality cost, generate the optimal service migration strategy using a deep Q-network model, and schedule the corresponding mobile edge computing nodes to perform service migration according to the optimal service migration strategy.

[0039] Specifically, the system takes minimizing the long-term accumulated comprehensive service quality cost as its decision objective. It calls a pre-trained Deep Q-Network (DQN) model to generate the optimal service migration strategy for the current moment. The Deep Q-Network model contains an evaluation network and a target network with the same structure but different parameter update frequencies. The model takes the complete global network state reconstructed in the previous steps as the input state space and outputs the action value function estimate corresponding to different candidate migration target nodes. Through continuous interaction with the network environment and experience replay mechanism, the network parameters are continuously optimized. Finally, the migration action that minimizes the long-term expected comprehensive cost is selected as the output strategy. The centralized controller then issues scheduling instructions to the corresponding mobile edge computing nodes to actually execute the service migration, thereby achieving intelligent, robust and forward-looking dynamic scheduling of computing tasks under imperfect telemetry conditions in industrial sites.

[0040] A deep Q-network (DQN) is used, where the reconstructed complete state tensor is input into the evaluation network, and the action value function is output. The agent extracts state transition samples through an experience replay mechanism. Update network parameters using the following loss function. : (4) in, Based on comprehensive cost Constructed instant rewards As a discount factor, These are the target network parameters.

[0041] The centralized controller periodically acquires new sparse observation tensors based on real-time feedback of system latency and load conditions. It continuously alternates between TT tensor completion and DQN model training.

[0042] The above describes the steps of the PON-MEC service migration method provided by this invention. As can be seen from the above description, the PON-MEC service migration method provided by this invention is applied to industrial passive optical networks and mobile edge computing PON-MEC systems where network state data is missing. It samples and processes multi-dimensional contextual attributes collected from the PON-MEC system according to a preset sparse network state observation model to form sparse network state observation information. The sparse network state observation model retains data at observable locations through a preset set of sampling locations, while unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information is deduced and completed to reconstruct the complete global network state. Based on the reconstructed complete global network state, the overall quality of service cost required for service migration between different mobile edge computing nodes is evaluated. With the goal of minimizing the long-term overall quality of service cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy. Therefore, this invention first obtains missing observation information based on a pre-defined sparse network state observation model, and then uses tensor column decomposition to extrapolate and complete the missing data, reconstructing a complete global network state, thereby eliminating blind spots in state awareness. Based on this, it evaluates the overall quality of service cost of service migration and generates an optimal migration strategy using a deep Q-network with the goal of minimizing the long-term overall cost. The synergy of these technical means enables the system to accurately perceive the overall network load and resource distribution even under harsh conditions such as limited telemetry and incomplete data in industrial settings, making forward-looking and stable migration decisions, significantly reducing average service latency and transmission jitter, and ensuring service continuity and deterministic transmission requirements when mobile terminals switch regions.

[0043] Based on the above embodiments, in this embodiment, step 200, based on tensor column decomposition, deduces and completes the missing data in the sparse network state observation information to reconstruct the complete global network state, including: Step 210: Decompose the high-dimensional tensor of the complete global network state to be reconstructed into a product of multiple third-order core tensors.

[0044] Step 220: Construct an optimization problem with the goal of minimizing the error between the sparse network state observation information and the reconstructed data at the preset sampling location set.

[0045] Step 230: Solve the optimization problem by iteratively updating the multiple third-order core tensors to deduce the missing data at unsampled locations and obtain the complete global network state.

[0046] Specifically, in step 210, the system abstracts the complete global network state to be reconstructed into a high-dimensional tensor structure. Each dimension of the high-dimensional tensor can correspond to the service type dimension, the optical network unit coverage area dimension, the user mobility mode dimension, and the edge computing node resource state dimension, etc. The high-dimensional tensor is approximated as a compact form of multiple smaller third-order core tensors multiplied sequentially along a specific pattern using tensor column decomposition technology.

[0047] Subsequently, in step 220, the system constructs a corresponding optimization problem with the goal of minimizing the deviation between the data values ​​reconstructed at the preset sampling location set and the actual collected sparse network state observation information. The preset sampling location set refers to those observable locations in the aforementioned sparse network state observation model that retain valid data due to successful telemetry. The essence of this optimization problem is to find a set of optimal core tensor parameters that minimize the reconstruction error at the known observation locations, under the premise of constraining the low-rank structure of each core tensor.

[0048] Finally, in step 230, the system uses an alternating iterative update method to solve each third-order core tensor one by one, thereby stably deducing the network state values ​​hidden at the originally unsampled locations under the condition of relying only on some observable data, and finally obtaining a complete and smooth global network state tensor, which provides a reliable environmental awareness basis for service quality cost assessment and migration decision in subsequent steps.

[0049] The PON-MEC service migration method provided in this embodiment decomposes the high-dimensional global network state tensor into a product of multiple third-order core tensors and iteratively solves the problem with the goal of minimizing the reconstruction error of observable locations. This method can accurately infer and complete the missing network state information under the condition of relying only on sparse and incomplete observation data, thereby providing a complete and reliable environmental perception basis for subsequent migration decisions.

[0050] Based on the above embodiments, in this embodiment, the deep Q-network model includes an evaluation network and a target network; Step 400 aims to minimize the long-term overall quality of service cost by generating an optimal service migration strategy using a deep Q-network model, including: Step 410: Input the complete global network state into the evaluation network and output the action value corresponding to each candidate migration action.

[0051] Step 420: By interacting with the network environment, collect state transition samples containing the current state, the migration action performed, the immediate reward obtained, and the next state, and store the state transition samples in the experience replay pool; wherein, the immediate reward is a negative value of the overall service quality cost.

[0052] Step 430: Extract target state transition samples from the experience replay pool, with the goal of minimizing the difference between the action value output by the evaluation network and the target action value calculated based on the target network, and update the parameters of the evaluation network.

[0053] Step 440: Based on the updated evaluation network, determine the migration action that minimizes the long-term expected overall service quality cost, and generate the optimal service migration strategy.

[0054] It should be noted that the deep Q-network model adopts a dual-network architecture, which includes an evaluation network and a target network with the same structure but different parameter update frequencies.

[0055] Specifically, in step 410, the system feeds the reconstructed complete global network state as the state input into the evaluation network. The evaluation network extracts and compresses the features of the high-dimensional state through multi-layer nonlinear transformation, and finally gives the action value function estimate corresponding to each candidate migration target node at the output layer.

[0056] Subsequently, in step 420, the system collects state transition samples for model training through continuous interaction with the actual network environment. Each complete sample record includes the current complete global network state before the transfer action is executed, the transfer action selected and actually executed according to the current policy, the immediate reward of the environment after the action is executed, and the complete global network state observed at the next moment after the action is executed. The value of the immediate reward is set to be the negative value of the overall service quality cost evaluated in step 300. This means that the shorter the latency and the lower the energy consumption caused by the transfer action, the larger the value of the immediate reward obtained, thereby guiding the agent to optimize decision-making behavior in the direction of reducing the overall cost. The collected state transition samples are uniformly stored in an experience replay pool with a fixed capacity to break the temporal correlation between consecutively sampled samples.

[0057] Next, in step 430, the system randomly selects a batch of target state transition samples from the experience replay pool for network parameter updates. For each sample, the next state is first input into the target network to obtain the maximum action value estimate of the next state, and then added to the immediate reward of the current sample to obtain the target action value. At the same time, the current state is input into the evaluation network to obtain the current action value output by the evaluation network. Then, a loss function is constructed with the goal of minimizing the difference between the evaluation network output value and the target action value.

[0058] Finally, in step 440, the system calculates the action value corresponding to each candidate migration action under the current complete global network state based on the evaluation network updated by the above iterations, and selects the one with the largest action value as the optimal migration action for the final output, thereby generating the optimal service migration strategy.

[0059] The PON-MEC service migration method provided in this embodiment adopts a dual-network deep Q-network architecture that includes an evaluation network and a target network. It takes the complete global network state as input, uses the negative value of the comprehensive service quality cost as an immediate reward, and uses the experience replay pool and the target network co-training mechanism to generate the optimal service migration strategy. This can effectively suppress value estimation oscillations and reduce the impact of temporal correlation of state transition samples on training stability.

[0060] Based on the above embodiments, in this embodiment, after scheduling the corresponding mobile edge computing node to perform service migration according to the optimal service migration strategy in step 400, the method further includes: Step 500: Update the sparse network state observation information and the deep Q-network model according to the dynamic changes in the network environment.

[0061] Step 500 specifically includes: Step 510: Obtain new multidimensional context attributes according to a preset period, and form updated sparse network state observation information based on the sparse network state observation model.

[0062] Step 520: Based on the updated sparse network state observation information, perform the tensor column decomposition again to infer and complete the missing data, and obtain the updated complete global network state.

[0063] Step 530: Based on the updated complete global network state, incrementally train the deep Q-network model to dynamically adapt to changes in the network environment.

[0064] Specifically, in step 510, the centralized controller periodically reacquires the multidimensional context attribute data of each node in the access system according to a preset acquisition cycle. This preset cycle can be flexibly configured according to the intensity of network dynamic changes and the processing capability of the centralized controller. For example, in the scenario of an automated production line where automated guided vehicles shuttle densely, it can be set to the level of hundreds of milliseconds. The newly acquired multidimensional context attributes are also filtered and processed by the sampling location set defined in the preset sparse network state observation model. Due to the continuous existence of electromagnetic interference and telemetry packet loss in the industrial field, there are still some missing data at unknown locations in the newly formed sparse network state observation information.

[0065] Subsequently, in step 520, the system takes the updated sparse network state observation information as input and calls the tensor column decomposition and deduction completion process described in step 200 again. By iteratively optimizing and solving each third-order core tensor, the system infers and fills in the network state values ​​of the positions that were not successfully sampled in this cycle, thereby obtaining the updated complete global network state that reflects the latest overall network situation.

[0066] Finally, in step 530, based on the previously trained model, the updated complete global network state is used as a new training sample to perform incremental training or online fine-tuning of the model. That is, based on the existing network weight parameters, several rounds of iterative optimization are performed with a small learning rate, so that the deep Q network model can smoothly adapt to dynamic environmental factors such as changes in network load distribution, user mobility pattern drift, and increases or decreases in edge node resources, avoiding model performance degradation and outdated decisions caused by non-stationary environmental changes.

[0067] The PON-MEC service migration method provided in this embodiment, through the closed-loop mechanism of periodic observation updates, tensor completion reconstruction, and incremental model training, can always maintain timely tracking and adaptive response to dynamic changes in the industrial field, ensuring the continuous optimization and stable reliability of the service migration strategy throughout its entire lifecycle.

[0068] The PON-MEC service migration method provided by the present invention will be described below in a complete embodiment.

[0069] In this embodiment, the network performance of the service migration method to be analyzed is the average service latency and latency jitter.

[0070] In this embodiment, an industrial passive optical network (PON) and mobile edge computing (MEC) system environment was established. The system deployed eight heterogeneous optical network units (ONUs), equipped with central processing units (CPUs) with frequencies ranging from 2.0 GHz to 22.0 GHz, serving 24 automated guided vehicles (AGVs). The AGVs' movement trajectories followed a Gaussian-Markov process with a characteristic correlation coefficient set to 0.7, and their speeds ranged from 5.56 m / s to 16.67 m / s with a standard deviation of 1.0 m / s. Real-world, multi-type tasks generated in the industrial environment (including security monitoring, video streaming, and control signaling) arrived according to a Poisson distribution, with an arrival rate of 5.8 tasks per user per second. The control and scheduling granularity of the system environment was set to 0.1 s. To realistically simulate the imperfect telemetry conditions in a harsh industrial environment, the system's random data loss rate was set to 10%. Under these simulation conditions, this embodiment compares and evaluates the scheduling method based on tensor columns and deep Q-networks (TT-DQN) proposed in this invention with the greedy Hungarian algorithm, particle swarm optimization (PSO) algorithm, and static benchmark algorithm that does not perform migration.

[0071] The system first senses and builds a sparse observation model, then reconstructs the global state through tensor column decomposition, then comprehensively evaluates the QoS cost, generates the optimal migration policy using DQN, and finally issues the policy and performs dynamic closed-loop updates of the system.

[0072] Figure 2 This is a schematic diagram of the industrial network system architecture based on PON-MEC and the TT-DQN algorithm execution framework provided by the present invention, as shown below. Figure 2 As shown, this embodiment provides a PON-MEC collaborative architecture for industrial automation scenarios and the execution environment for the above migration algorithm. Figure 2 The upper part of the diagram shows the Industrial Network topology, which consists of a centralized control center and multiple distributed Optical Network Units (ONUs). The ONUs provide low-latency access services to manufacturing workshops, on-site monitoring areas, and other locations via wireless networks. Figure 2 The lower half of the diagram shows the specific mapping of the Algorithm Layer: the incomplete state features are constructed as sparse observation tensors, and the complete reconstructed state is output through low-rank inference by the TT recovery module. Finally, the reconstructed state is input into the DQN agent to output the transition instructions.

[0073] Figure 3 This is a schematic diagram comparing the average latency performance under different network load conditions provided by the present invention, such as... Figure 3 As shown, to demonstrate the accuracy of the method proposed in this invention, the comparison results of average latency under different normalized loads are presented with and without tensor completion mechanism, at a network telemetry data missing rate of 10%. Figure 3 As shown, as the normalized load gradually increases from 0.1 to 1.0, the TT-DQN strategy of this invention maintains the lowest average latency. Especially under full load conditions with a normalized load of 1.0, the average latency of the method of this invention is effectively controlled at approximately 28 ms, while the average latency of the PSO algorithm combined with tensor completion and the Hungarian algorithm both reach over 45 ms; without the static baseline algorithm for migration scheduling, the average latency rises sharply to over 140 ms. This demonstrates the significant advantages of this invention in overcoming data loss and significantly reducing average computation and transmission latency under high load.

[0074] Figure 4 This is a schematic diagram comparing latency jitter performance under different network load conditions provided by the present invention, such as... Figure 4This paper presents a comparative evaluation of the latency jitter performance of the proposed strategy and benchmark algorithms under different normalized load conditions. In imperfect observation environments with missing data, latency jitter directly reflects the stability of the system's service quality. Experimental results show that the TT-DQN strategy of this invention achieves extremely low latency jitter across the entire load variation range. Specifically, when the normalized load is 1.0, the latency jitter of the proposed method is only about 5 ms; in contrast, the jitter of PSO combined with tensor completion and the Hungarian algorithm is between 15 ms and 20 ms, while the latency jitter of the static benchmark algorithm exceeds 55 ms. This fully demonstrates that, with accurate global state reconstruction and the long-term decision-making mechanism of DQN, this invention can effectively avoid frequent or ineffective scheduling switches when facing imperfect telemetry data in industrial settings, thereby meeting the stringent time determinism requirements of high-precision industrial tasks.

[0075] The PON-MEC service migration method provided in this embodiment combines tensor column decomposition technology with deep reinforcement learning algorithms. Applied to industrial PON-MEC systems with missing network state data, it effectively supports concurrent computing and seamless migration of various types of industrial services. Through a collaborative mechanism of accurate global state reconstruction and intelligent long-term optimization, this method significantly improves the system's scheduling performance, meeting the stringent requirements of time-sensitive services for low latency and low jitter.

[0076] The PON-MEC service migration apparatus provided by the present invention is described below. The PON-MEC service migration apparatus described below and the PON-MEC service migration method described above can be referred to in correspondence.

[0077] Figure 5 This is a schematic diagram of the structure of the PON-MEC service migration device provided by the present invention, as shown below. Figure 5 As shown, the PON-MEC service migration device provided by this invention is applied to an industrial passive optical network and mobile edge computing PON-MEC system where network status data is missing. The device includes: The sampling module 501 is used to sample and process the multi-dimensional context attributes collected from the PON-MEC system according to the preset sparse network state observation model to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. Reconstruction module 502 is used to infer and complete the missing data in the sparse network state observation information based on tensor column decomposition, and reconstruct the complete global network state. Evaluation module 503 is used to evaluate the overall quality of service cost required for migrating services between different mobile edge computing nodes based on the reconstructed complete global network state. The migration module 504 is used to generate an optimal service migration strategy using a deep Q-network model with the goal of minimizing the long-term overall service quality cost, and to schedule the corresponding mobile edge computing nodes to perform service migration according to the optimal service migration strategy.

[0078] The PON-MEC service migration device provided by this invention is applied to industrial passive optical networks and mobile edge computing PON-MEC systems where network state data is missing. It samples multi-dimensional contextual attributes collected from the PON-MEC system according to a preset sparse network state observation model to form sparse network state observation information. The sparse network state observation model retains data at observable locations through a preset set of sampling locations, while unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information is deduced and completed to reconstruct a complete global network state. Based on the reconstructed complete global network state, the overall quality of service cost required for service migration between different mobile edge computing nodes is evaluated. With the goal of minimizing the long-term overall quality of service cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy. Therefore, this invention first obtains missing observation information based on a pre-defined sparse network state observation model, and then uses tensor column decomposition to extrapolate and complete the missing data, reconstructing a complete global network state, thereby eliminating blind spots in state awareness. Based on this, it evaluates the overall quality of service cost of service migration and generates an optimal migration strategy using a deep Q-network with the goal of minimizing the long-term overall cost. The synergy of these technical means enables the system to accurately perceive the overall network load and resource distribution even under harsh conditions such as limited telemetry and incomplete data in industrial settings, making forward-looking and stable migration decisions, significantly reducing average service latency and transmission jitter, and ensuring service continuity and deterministic transmission requirements when mobile terminals switch regions.

[0079] Based on the above embodiments, in this embodiment, the multidimensional context attributes include at least one of service type, optical network unit coverage area, user mobility mode, and central processing unit load status of mobile edge computing nodes; the preset sampling location set is predetermined based on network telemetry capabilities and signaling overhead constraints.

[0080] Based on the above embodiments, in this embodiment, the reconstruction module 502 is specifically used for: The high-dimensional tensor of the complete global network state to be reconstructed is decomposed into a product of multiple third-order core tensors. An optimization problem is constructed with the objective of minimizing the error between the sparse network state observation information and the reconstructed data at the preset sampling location set; By iteratively updating the multiple third-order core tensors, the optimization problem is solved to deduce the missing data at unsampled locations and obtain the complete global network state.

[0081] Based on the above embodiments, in this embodiment, the overall quality of service cost is determined by a weighted sum of service latency and energy consumption; wherein, the service latency includes wireless transmission latency, the computation and processing latency of the target mobile edge computing node, and the fiber optic backhaul latency of the passive optical network system.

[0082] Based on the above embodiments, in this embodiment, the deep Q-network model includes an evaluation network and a target network; The migration module 504 is specifically used for: The complete global network state is input into the evaluation network, and the action value corresponding to each candidate migration action is output. By interacting with the network environment, state transition samples containing the current state, the migration action performed, the immediate reward obtained, and the next state are collected, and the state transition samples are stored in the experience replay pool; wherein, the immediate reward is the negative value of the overall service quality cost; The parameters of the evaluation network are updated by extracting target state transition samples from the experience replay pool and minimizing the difference between the action value output by the evaluation network and the target action value calculated based on the target network. Based on the updated evaluation network, the migration action that minimizes the long-term expected overall service quality cost is determined, and the optimal service migration strategy is generated.

[0083] Based on the above embodiments, in this embodiment, the device further includes an update module, specifically used for: After the mobile edge computing nodes corresponding to the optimal service migration strategy are scheduled to perform service migration, The sparse network state observation information and the deep Q-network model are updated according to the dynamic changes in the network environment.

[0084] Based on the above embodiments, in this embodiment, the updating module is specifically used for: New multidimensional context attributes are acquired according to a preset period, and updated sparse network state observation information is formed based on the sparse network state observation model. Based on the updated sparse network state observation information, the tensor column decomposition is performed again to infer and complete the missing data, and the updated complete global network state is obtained. Based on the updated complete global network state, the deep Q-network model is incrementally trained to dynamically adapt to changes in the network environment.

[0085] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device can be a robot or other electronic device. This electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communication bus 640. The processor 610, communications interface 620, and memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a PON-MEC service migration method, applied to industrial passive optical networks and mobile edge computing PON-MEC systems where network status data is missing. The method includes: According to the preset sparse network state observation model, the multi-dimensional context attributes collected from the PON-MEC system are sampled and processed to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information are deduced and completed to reconstruct the complete global network state. Based on the reconstructed complete global network state, assess the overall quality of service cost required for migrating services between different mobile edge computing nodes; With the goal of minimizing the long-term overall service quality cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy.

[0086] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in at least one embodiment of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the PON-MEC service migration methods provided by the above methods, and apply them to industrial passive optical networks and mobile edge computing PON-MEC systems where network status data is missing. The method includes: According to the preset sparse network state observation model, the multi-dimensional context attributes collected from the PON-MEC system are sampled and processed to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information are deduced and completed to reconstruct the complete global network state. Based on the reconstructed complete global network state, assess the overall quality of service cost required for migrating services between different mobile edge computing nodes; With the goal of minimizing the long-term overall service quality cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy.

[0088] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the PON-MEC service migration methods provided by the above methods, applied to industrial passive optical networks and mobile edge computing PON-MEC systems where network state data is missing, the method comprising: According to the preset sparse network state observation model, the multi-dimensional context attributes collected from the PON-MEC system are sampled and processed to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information are deduced and completed to reconstruct the complete global network state. Based on the reconstructed complete global network state, assess the overall quality of service cost required for migrating services between different mobile edge computing nodes; With the goal of minimizing the long-term overall service quality cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy.

[0089] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0090] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A PON-MEC service migration method, characterized in that, The method, applied to industrial passive optical networks and mobile edge computing (PON-MEC) systems where network status data is missing, includes: According to the preset sparse network state observation model, the multi-dimensional context attributes collected from the PON-MEC system are sampled and processed to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. Based on tensor column decomposition, the missing data in the sparse network state observation information are deduced and completed to reconstruct the complete global network state. Based on the reconstructed complete global network state, assess the overall quality of service cost required for migrating services between different mobile edge computing nodes; With the goal of minimizing the long-term overall service quality cost, an optimal service migration strategy is generated using a deep Q-network model, and the corresponding mobile edge computing nodes are scheduled to perform service migration according to the optimal service migration strategy.

2. The PON-MEC service migration method according to claim 1, characterized in that, The multidimensional context attributes include at least one of the following: service type, optical network unit coverage area, user mobility mode, and central processing unit load status of mobile edge computing nodes; the preset sampling location set is predetermined based on network telemetry capabilities and signaling overhead constraints.

3. The PON-MEC service migration method according to claim 1, characterized in that, The process of extrapolating and completing missing data in the sparse network state observation information based on tensor column decomposition to reconstruct the complete global network state includes: The high-dimensional tensor of the complete global network state to be reconstructed is decomposed into a product of multiple third-order core tensors. An optimization problem is constructed with the objective of minimizing the error between the sparse network state observation information and the reconstructed data at the preset sampling location set; By iteratively updating the multiple third-order core tensors, the optimization problem is solved to deduce the missing data at unsampled locations and obtain the complete global network state.

4. The PON-MEC service migration method according to claim 1, characterized in that, The overall quality of service cost is determined by a weighted sum of service latency and energy consumption; wherein, the service latency includes wireless transmission latency, computation and processing latency of the target mobile edge computing node, and fiber optic backhaul latency of the passive optical network system.

5. The PON-MEC service migration method according to claim 1, characterized in that, The deep Q-network model includes an evaluation network and a target network; The method of generating an optimal service migration strategy using a deep Q-network model with the objective of minimizing the long-term overall service quality cost includes: The complete global network state is input into the evaluation network, and the action value corresponding to each candidate migration action is output. By interacting with the network environment, state transition samples containing the current state, the migration action performed, the immediate reward obtained, and the next state are collected, and the state transition samples are stored in the experience replay pool; wherein, the immediate reward is the negative value of the overall service quality cost; The parameters of the evaluation network are updated by extracting target state transition samples from the experience replay pool and minimizing the difference between the action value output by the evaluation network and the target action value calculated based on the target network. Based on the updated evaluation network, the migration action that minimizes the long-term expected overall service quality cost is determined, and the optimal service migration strategy is generated.

6. The PON-MEC service migration method according to any one of claims 1-5, characterized in that, After scheduling the corresponding mobile edge computing node to perform service migration according to the optimal service migration strategy, the method further includes: The sparse network state observation information and the deep Q-network model are updated according to the dynamic changes in the network environment.

7. The PON-MEC service migration method according to claim 6, characterized in that, The step of updating the sparse network state observation information and the deep Q-network model according to dynamic changes in the network environment includes: New multidimensional context attributes are acquired according to a preset period, and updated sparse network state observation information is formed based on the sparse network state observation model. Based on the updated sparse network state observation information, the tensor column decomposition is performed again to infer and complete the missing data, and the updated complete global network state is obtained. Based on the updated complete global network state, the deep Q-network model is incrementally trained to dynamically adapt to changes in the network environment.

8. A PON-MEC service migration device, characterized in that, The device, applied in industrial passive optical networks and mobile edge computing (PON-MEC) systems where network status data is missing, comprises: The sampling module is used to sample and process the multi-dimensional context attributes collected from the PON-MEC system according to the preset sparse network state observation model to form sparse network state observation information; wherein, the sparse network state observation model retains the data at observable locations through a preset set of sampling locations, and the unsampled locations show missing data. The reconstruction module is used to infer and complete the missing data in the sparse network state observation information based on tensor column decomposition, and reconstruct the complete global network state. The evaluation module is used to evaluate the overall quality of service cost required for migrating services between different mobile edge computing nodes based on the reconstructed complete global network state. The migration module is used to generate an optimal service migration strategy using a deep Q-network model with the goal of minimizing the long-term overall service quality cost, and to schedule the corresponding mobile edge computing nodes to perform service migration according to the optimal service migration strategy.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the PON-MEC service migration method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the PON-MEC service migration method as described in any one of claims 1 to 7.

Images