MEC task unloading and resource allocation optimization method for industrial Internet of Things

Abstract

The embodiment of the invention provides an MEC task unloading and resource allocation optimization method oriented to the industrial Internet of Things. The method comprises the following steps: acquiring an MEC task from terminal equipment; according to a pre-established unloading decision algorithm, determining the MEC task as an auction candidate task or a local execution task; according to a pre-established auction algorithm, determining the auction candidate task as an auction winning task or a re-decision task; and allocating the auction winning task to the corresponding edge node for execution. According to the embodiment of the invention, the unloading decision algorithm based on the improved DQN is provided, the load balance and task time delay constraint of the edge node are fully considered, the cooperative calculation of the terminal equipment and the edge node is realized, and the calculation energy efficiency of the Internet of Things is maximized on the premise of meeting the time delay limitation; a cyclic combined auction algorithm is provided in combination with an economics theory to describe a resource allocation process in an edge computing scene, and social welfare is optimized.

Description

Technical Field

[0001] Embodiments of the present disclosure belong to the technical field of MEC task offloading and resource allocation optimization, and particularly relate to an MEC task offloading and resource allocation optimization method for industrial Internet of Things. Background Art

[0002] As an application of the Internet of Things in the industrial field, the industrial Internet of Things combines various sensors, monitoring devices, automation devices, and terminal devices with the Internet through advanced communication and network technologies, making the industrial system operate more efficiently and intelligently. In response to the rapid growth of task quantity and computing volume in the industrial Internet of Things, Mobile Edge Computing (MEC) can timely process computationally intensive and latency-sensitive tasks and provide low-latency, energy-saving, and secure services by deploying resources at the network edge close to terminal devices, and has become an effective means to improve the stable operation, real-time analysis, and control capabilities of the industrial Internet of Things.

[0003] Due to cost and feasibility factors, the resources of the edge network are limited, and algorithms that can effectively manage task load (computing offloading) and schedule resources (resource allocation) are required. Through computing offloading, deciding which edge server a mobile device offloads its tasks to can reduce the computing burden of the terminal device and extend the battery life of the device, but it will cause additional transmission delay and power consumption. And resource allocation is crucial for computing offloading in the MEC system. An effective resource allocation strategy can enable edge nodes to accommodate as many computing offloading requests from different terminal devices as possible, thereby improving the overall performance of the system.

[0004] Existing heuristic algorithms or computing offloading strategies based on mathematical programming require sufficient prior knowledge to identify complex relationships between interdependent parameters and cannot make adaptive decisions quickly. And existing resource allocation research only considers the needs of terminal devices and system performance, while ignoring the energy consumption benefits of edge servers. The stakeholders in the industrial Internet of Things tend to be selfish when requesting resources, that is, they pursue the maximization of their own energy efficiency while ignoring the overall energy efficiency; at the same time, with the increase in the number of terminal devices, edge nodes with limited resources cannot complete all computing tasks and must selectively allocate resources. Summary of the Invention

[0005] Embodiments of the present disclosure aim to at least solve one of the technical problems existing in the prior art, and provide an MEC task offloading and resource allocation optimization method for industrial Internet of Things. The Internet of Things includes a terminal device and an edge node that are electrically connected to each other. The method includes:

[0006] Obtaining MEC tasks from the terminal device;

[0007] ​

[0008] According to the pre-established offloading decision algorithm, determine the MEC task as a bidding candidate task or a local execution task; wherein, the offloading decision algorithm is pre-established by training a deep Q-network based on the computing energy efficiency of the Internet of Things;

[0009] According to the pre-established auction algorithm, determine the bidding candidate task as a winning bidding task or a re-decision task; wherein, the auction algorithm is pre-established based on the optimal matching algorithm, the heuristic algorithm, and the Vickrey auction rule;

[0010] Allocate the winning bidding task to the corresponding edge node for execution.

[0011] Furthermore, the offloading decision algorithm is pre-established through the following steps:

[0012] Calculate the computing energy efficiency of the Internet of Things based on the computing latency and computing energy consumption of the terminal device and the edge node;

[0013] Define the optimization problem of the computing energy efficiency and transform the optimization problem into a Markov model;

[0014] Train the deep Q-network in combination with the Markov model and the Epsilon-Greedy algorithm to obtain the offloading decision algorithm.

[0015] Furthermore, the computing latency of the terminal device is:

[0016]

[0017] In the formula, is the offloading decision variable of task i k ; is the computational amount of task i k ; is the CPU resource ratio allocated by the terminal device to task i k ; f k is the local computing ability of terminal device k;

[0018] The computing energy consumption of task i k on the terminal device is:

[0019]

[0020] In the formula, is the computing power of the terminal device;

[0021] The computing latency of the edge node j is:

[0022] <\

[0023] In the formula, is the offloading decision variable for task i k , is the computational complexity of task i k , is the proportion of computing resources allocated by edge node j to task i k , is the computing power of edge node j;

[0024] The transmission delay from the terminal device to edge node j for task i is: k

[0025]

[0026] In the formula, is the link length for task i k uploaded to edge node j, is the upload rate for task i k uploaded to edge node j;

[0027] The transmission energy consumption from the terminal device to edge node j for task i is: k

[0028]

[0029] In the formula, p k is the transmission power for uploading task i k to edge node j;

[0030] The total execution delay from the terminal device to edge node j for task i is: k

[0031]

[0032] The computing energy efficiency of the Internet of Things is:

[0033]

[0034] In the formula, B total is the total number of computing bits completed by the Internet of Things within the duration, and E total is the total energy consumption of the Internet of Things.

[0035] Furthermore, the optimization problem is:

[0036] maxU sys

[0037]

[0038]

[0039] In the formula, the objective function maxU​​​sys That is, to maximize the computing energy efficiency of the Internet of Things, is the maximum tolerable delay of the task, is the bandwidth ratio allocated by edge node j to task i k Further, the auction algorithm includes:

[0040] Define the heuristic factor of the edge node;

[0041] Use the optimal matching algorithm to select the edge node with the highest heuristic factor, and determine the corresponding auction candidate task as the winning auction task;

[0042] According to the Vickrey auction rule, determine the payment price of the terminal device corresponding to the winning auction task.

[0043] An MEC task offloading and resource allocation optimization method for industrial Internet of Things in an embodiment of the present disclosure, by proposing an offloading decision algorithm based on improved DQN, fully considering the load balance of edge nodes and task delay constraints, realizes the collaborative computing between terminal devices and edge nodes, and maximizes the computing energy efficiency of the Internet of Things on the premise of meeting the delay limit; a cyclic combinatorial auction algorithm proposed in combination with economic theory is used to describe the resource allocation process in the edge computing scenario and optimize social welfare.

[0044] BRIEF DESCRIPTION OF THE DRAWINGS BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic flowchart of an MEC task offloading and resource allocation optimization method for industrial Internet of Things in an embodiment of the present disclosure;

[0046] Figure 2 is a schematic structural diagram of an MEC task offloading and resource allocation optimization system for industrial Internet of Things in another embodiment of the present disclosure;

[0047] Figure 3 is a schematic structural diagram of an electronic device in another embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] Next, the technical solutions in the embodiments of the present disclosure will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0049] In addition, the described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present disclosure. However, those skilled in the art will realize that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0050] The flowchart shown in the accompanying drawings is only an exemplary illustration, and does not necessarily include all contents and operations / steps, nor is it necessary to be executed in the described order. For example, some operations / steps can be decomposed, while some operations / steps can be combined or partially combined, so the actual execution order may change according to the actual situation.

[0051] It should be understood that although terms such as first, second, and third may be used in the present disclosure to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another. Therefore, the first component discussed below can be referred to as the second component without departing from the teachings of the concepts of the present disclosure. As used in the present disclosure, the term "and / or" includes any one of the associated listed items and all combinations of one or more of them.

[0052] Those skilled in the art can understand that the drawings are only schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily essential for implementing the present disclosure, so they cannot be used to limit the protection scope of the present disclosure.

[0053] An embodiment of the present disclosure provides an MEC task offloading and resource allocation optimization method for the industrial Internet of Things, based on a cloud-edge-end three-layer edge computing offloading and resource allocation model in an industrial Internet of Things. Among them, the cloud layer has rich communication resources and computing resources. The cloud layer is responsible for the training and update of the task offloading decision model. At the same time, as a trusted third party, the cloud layer can act as an auctioneer, collect resource requirements from terminal devices and resource usage situations of edge server nodes, complete the auction process, and make resource allocation decisions through a dedicated control channel. The edge layer consists of J edge nodes, and each edge node includes a small base station and an edge server. Edge node j has certain communication and parallel computing capabilities. The computing power of the edge server is characterized by the CPU rate, that is, the maximum number of cycles that the CPU can run per second. The terminal device layer consists of K terminal devices, such as sensors, etc. The terminal devices can generate computing tasks and receive the returned computing results. The i-th computing task of terminal device K is defined as i k ∈{1,2,...,I}, for the task data size, is the computational workload of the task, that is, the number of CPU cycles required for the computing task, is the maximum tolerable delay of the task. Tasks arrive randomly and the characteristic parameters follow a Poisson distribution. The terminal device uses the training model sent by the cloud to generate migration decisions and perform data transmission and computing.

[0054] Define a discrete time slot model \(t\in\{1,2,\cdots,T\}\), and the duration of each time slot is \(\tau\). The terminal device can compute tasks locally or migrate the tasks to the edge node. The task offloading decision variable indicates that the task is executed locally, indicates that the task is offloaded to the edge node for execution.

[0055] As Figure 1 shown, an MEC task offloading and resource allocation optimization method for industrial Internet of Things in an embodiment of the present disclosure includes:

[0056] Step S1, obtain the MEC task from the terminal device.

[0057] Specifically, the MEC computing task is generated by the terminal device at the terminal device layer. When the local computing resources of the terminal device are insufficient, it can be offloaded to the edge node device with rich computing resources for computing through MEC technology.

[0058] Step S2, determine the MEC task as a bidding candidate task or a local execution task according to a pre-established offloading decision algorithm; wherein, the offloading decision algorithm is pre-established by training the deep Q network based on the computing energy efficiency of the Internet of Things.

[0059] Specifically, the terminal device decides whether to keep the MEC task for local execution or offload it to the edge node for processing according to the pre-trained offloading decision algorithm.

[0060] Regarding the offloading decision algorithm, first establish a communication model. When the terminal device selects to offload the task to the edge server, there will be a transmission delay. Assume that multi-terminal device interference is eliminated through orthogonal frequency division multiplexing technology. For task \(i\) k The uplink rate of migrating to edge node \(j\) is as follows:

[0061]

[0062] In the formula, \(B\) j represents the base station bandwidth of edge node \(j\), \(p\) k represents the transmission power of terminal device \(k\) to upload data, \(g\) k represents the channel gain of terminal device \(k\) in the wireless channel, and \(N_0\) represents the Gaussian white noise power spectral density, Indicates the bandwidth occupancy ratio of the task assigned by edge node j to terminal device k.

[0063] Then, in the migration calculation, the transmission delay of uploading task i k from the terminal device to edge node j can be expressed as:

[0064]

[0065] Wherein, is the link length of uploading task i k to edge node j, is the upload rate of uploading task i k to edge node j.

[0066] According to the transmission delay, the transmission energy consumption of uploading task i k from the terminal device to edge node j is:

[0067]

[0068] Wherein, p k is the transmission power of uploading task i k to edge node j. Since the size of the calculation result is much smaller than the size of the input task, the downlink transmission delay and energy consumption are ignored.

[0069] Subsequently, the calculation models for the local execution of MEC tasks on the terminal device and the execution on the edge node are established respectively. When the task is executed locally, the local execution delay is related to the processing capacity of the local CPU. Therefore, the local calculation delay of the task is:

[0070]

[0071] Wherein, is the offloading decision variable of task i k , is the amount of calculation of task i k , is the CPU resource occupancy ratio assigned by the terminal device to task i k , f k is the local calculation capacity of terminal device k.

[0072] Then, the calculation energy consumption of task i k on the terminal device is:

[0073]

[0074] Wherein, is the calculation power of the terminal device.

[0075] The tasks unloaded to the edge nodes are computed using the CPU resources allocated by the edge servers, and the terminal device task i k The computing delay on the edge node j is:

[0076]

[0077] In the formula, is the offloading decision variable of task i k ; is the amount of computation of task i k ; is the proportion of computing resources allocated by the edge node j to task i k ; is the computing power of the edge node j

[0078] The total execution delay of the terminal device task i k migrated to the edge node j can be obtained as:

[0079]

[0080] The offloading decision algorithm of the embodiments of the present disclosure aims to maximize the computing energy efficiency of the Internet of Things system. The computing energy efficiency is an index for evaluating the system performance, defined as the ratio of the total number of computed bits in the system over a period of time to the energy consumption. The computing energy efficiency of the entire Internet of Things including the terminal devices and the edge nodes is:

[0081]

[0082] In the formula, B total is the total number of computed bits completed by the Internet of Things during the duration, and E total is the total energy consumption of the Internet of Things

[0083] The specific optimization problem is constructed as follows:

[0084] maxU sys

[0085]

[0086] In the above optimization problem, the objective function maxU sys is to maximize the computing energy efficiency of the Internet of Things; Equation means that regardless of whether the task selects local computing or edge node computing, its delay cannot be greater than the maximum tolerable delay of the task; Equation means that in a certain time slot, the total computing resources allocated by the edge node to all tasks cannot exceed its computing power; Equation means that in the time slot, the total bandwidth resources allocated by the edge node to all tasks cannot exceed its bandwidth resources. Equation means the value constraint of the task offloading decision variable

[0087] In the computing offloading decision-making stage, the above optimization problem is transformed into a Markov model, thereby transforming the optimization problem into a Markov Decision Process (MDP) within a finite time horizon.

[0088] MDP is composed of a five-tuple (S, A, P, R, γ), where S represents the state space, A represents the action space, P represents the state transition probability, R represents the reward function, and γ is the discount factor, which is used to adjust the influence of short-term and long-term rewards on the agent. The specific description is as follows:

[0089] At the beginning of each time slot, the agent deployed at the terminal device collects information on the current local environment to form the state space. The state space includes the channel resource occupancy rate of each edge node in the system, the number of tasks being transmitted in the channel and the remaining transmission time, the computing resource occupancy rate, the number of tasks being computed and the remaining execution time, and the characteristic information of the current task. The state space is defined as s(t).

[0090] Subsequently, the task offloading algorithm based on the improved DQN is executed. First, the parameters of the Q neural network and the experience replay set are initialized. The core idea of DQN is to guide the agent's actions through rewards based on the observation of the system state. DQN uses the neural network Q(s, a, w) to approximate the action value function, where w represents the parameters of the neural network.

[0091] The agent observes the environmental state s(t) and selects an action a(t) to execute according to the policy. In this embodiment, the Epsilon-Greedy algorithm is used as the policy for action selection: that is, a positive number ε (ε < 1) is determined as the probability of randomly selecting an action, and the remaining probability of 1 - ε is used to select the action with the largest value obtained in Q(s, a, w).

[0092] Execute the action a(t) in the environment to obtain a new state s(t + 1) and a reward r(t). Once an action is taken based on the observed state, the agent can obtain a reward and enter the next state. In the training stage, the agent iteratively adjusts the policy according to the obtained reward until convergence. The reward function is usually related to the optimization objective. Therefore, the reward function is defined as the ratio of the number of bits of this task to the energy consumption (computing energy efficiency). Considering the latency requirement of the task, when the task duration exceeds its maximum tolerable latency, the task is regarded as a failure, and the terminal device will receive a penalty. The penalty is defined as the negative value of the absolute value of the current reward. The observed state transition sequence (s(t), a(t), r(t), s(t + 1)) is stored in the experience pool.

[0093] Randomly select a small batch (W) of samples for each training, calculate and optimize the loss function, and update the neural network parameters in the Q network until convergence:

[0094] 1. Use the experience replay method to solve the problem of local optimal solutions caused by sample similarity. The target value is calculated as follows during each training:

[0095]

[0096] where γ ranges from (0, 1] and is used to adjust the influence of future rewards on the current state;

[0097] 2. Use the L2 mean square error loss function to calculate the gap between and;

[0098] 3. Update the neural network parameters using the gradient descent method to reduce the loss function value and improve the accuracy of the neural network in fitting the action value function. The parameter update speed is controlled by the learning rate α.

[0099] Thus, the offloading decision algorithm is trained and established. The terminal device uses the trained offloading decision algorithm to make MEC task offloading decisions, adds the tasks determined to be offloaded to the auction candidate list of the selected edge node, that is, determines them as auction candidate tasks. The auction candidate tasks planned to be migrated to the edge node for execution will enter the auction phase.

[0100] Deep Reinforcement Learning (DRL) combines the advantages of deep learning and reinforcement learning and can search for asymptotically optimal solutions in a time-varying environment. Without any prior knowledge, DRL uses an agent to obtain the hidden features of the environment and learns the optimal policy through repeated interactions in a specific environment. This feature makes DRL show its special potential when designing computational offloading schemes in dynamic systems.

[0101] Compared with methods such as heuristic algorithms or strategies based on mathematical programming, the deep reinforcement learning algorithm adopted in this embodiment has a powerful self-learning ability. It can identify the complex relationships between interdependent parameters in the absence of prior knowledge, make adaptive decisions quickly, and is widely used to solve large-scale dynamic optimization and complex cooperation and game problems. The offloading decision algorithm proposed in this embodiment takes into account factors such as edge node load balancing and task queuing delay, can adaptively adjust the task migration strategy in a time-varying environment, and select the optimal discretized resource request strategy.

[0102] Step S3: According to the pre-established auction algorithm, determine the auction candidate task as an auction-winning task or a re-decision task; wherein, the auction algorithm is pre-established based on the optimal matching algorithm, heuristic algorithm, and Vickrey auction rule.

[0103] Specifically, based on its resource capacity, current resource demand status, and the bids of the terminal devices, the edge node selects the winning tasks through an auction mechanism to form a matching auction pair. The tasks that win the auction will prepare for the offloading operation; while the tasks that fail to win the auction need to adjust the bids of their terminal devices and return to the previous step S2 for offloading decision-making.

[0104] The edge node and the terminal device are regarded as the seller and the buyer of computing resources and channel resources respectively. A trusted third party such as a cloud server acts as the auctioneer, receives the bidding requests from the buyers, and collects the status information from the sellers.

[0105] Design the bidding strategy of the terminal device, and define the service cost and social welfare of the edge node:

[0106] 1. The buyers have different valuations for the sellers' services. The terminal device tends to offload computationally intensive tasks to edge nodes with rich computing resources, and offload energy consumption sensitive tasks to edge nodes with rich bandwidth resources. The bids submitted by the terminal device reflect the terminal device's consideration of valuation and budget. In this embodiment, it is assumed that the bidding price of the terminal device is real, that is, the bidding price is equivalent to the real valuation of the resources by the terminal device.

[0107] 2. The service cost of the edge node consists of three parts, including the depreciation cost per unit service time, the electricity cost paid per unit energy consumption, and the economic cost per unit bandwidth.

[0108] 3. is the buyer i k The bidding price submitted for the seller j, is the service cost for the seller j to provide resources for the buyer i k The social welfare is defined as the true value of all buyers minus the total service cost.

[0109] Based on the above bidding and quoting strategies, this embodiment designs a cyclic combinatorial auction algorithm. First, determine the auction winner according to the proposed matching mechanism. The buyer (task) i k can be matched with at most one seller (edge node) j, while the seller with limited CPU and channel resources can serve multiple buyers. Therefore, the buyer-seller matching problem can be regarded as a two-resource generalized assignment problem, which is an NP-hard problem. The present invention combines the sub-steps of the optimal matching algorithm and the heuristic algorithm to maximize the social welfare.

[0110] The specific steps include:

[0111] 1. Define the heuristic factor. Define the set of indices of the unmatched buyers as M, and the set of indices of the sellers who can provide available resources for the buyer i k as denote The heuristic factor is defined as:

[0112]

[0113] 2. Simplify the problem using the optimal matching algorithm. Ignoring the constraint that a buyer can only be matched with one seller, the buyer-seller matching problem can be decomposed into J sub-problems. The j-th sub-problem is:

[0114]

[0115] Each sub-problem can be regarded as a 0-1 knapsack problem. Each sub-problem can be solved in polynomial time using the dynamic programming method to obtain the solution X'. If there is exactly one seller j in the solution X' that matches the buyer i k , a buyer-seller matching pair is formed, and the buyer is removed from the auction candidate list.

[0116] 3. Apply the heuristic buyer-seller matching algorithm to solve the remaining problems. The auctioneer repeatedly executes the following steps: (1) Initialize the list I' of buyers to be auctioned in this stage. For all buyers i k ∈ I', calculate whether there is a seller who can provide available resources for it. If not, remove i k from I'; (2) Select the buyer with the highest heuristic factor in the set ; (3) Select the corresponding seller to form a buyer-seller matching pair and update the remaining available resources of the seller. Through this step, the infeasible solution is transformed into a feasible approximate optimal solution.

[0117] Finally, according to the Vickrey auction rule, determine the price that the winning terminal device should finally pay for its task:

[0118] In the pricing stage, the final bid of the task should be independent of its own bid. The serial number of the terminal device task in the descending bid queue is represented by l. According to the Vickrey auction rule, assuming that the task wins in a certain round of auction at the edge node j, the unit resource price that it should finally pay to the edge node is the higher value between the bids of the tasks behind it and the asking price of the node, that is

[0119] The winning tasks of the successful auction are ready to be offloaded. According to the matching result and the payment plan, the terminal device migrates the tasks to the edge nodes and makes payments. The tasks that fail the auction re-enter the previous step S2 for offloading decisions.

[0120] The Vickrey auction theory is a classic mechanism design model for resource allocation in economics and is widely applied to other fields involving resource competition. Existing auction mechanisms mainly focus on cloud computing resource auctions. Due to the heterogeneity of edge server resources, the transaction and game processes of each component of the edge system are more complex. Therefore, it is necessary to research and design an auction algorithm for resource allocation in the edge computing scenario to incentivize mobile terminal devices and edge servers to participate in offloading service transactions.

[0121] Existing auction research only considers the needs of terminal devices and system performance, while ignoring the benefits of edge servers. In fact, the stakeholders in the industrial Internet of Things tend to be selfish when requesting resources, that is, they pursue the maximization of their own interests while ignoring the overall interests. At the same time, as the number of terminal devices increases, the edge nodes with limited resources cannot complete all computing tasks and must selectively allocate resources. The auction algorithm proposed in this embodiment uses a multi-round iterative auction algorithm mechanism for resource allocation, which is applicable to the edge computing system and can improve the profit of edge nodes and resource utilization.

[0122] Step S4: Allocate the winning bid tasks to the corresponding edge nodes for execution.

[0123] Specifically, the base station allocates appropriate bandwidth resources for the winning bid tasks according to the requirements. Subsequently, the tasks are uploaded to the corresponding edge nodes through the wireless access network or the cellular mobile network and wait to be processed when the computing resources are idle. Once the computing resources of the edge node are idle, it will allocate the necessary computing resources for the waiting tasks to ensure that the tasks can be executed smoothly. After the tasks are executed, the terminal device pays the corresponding fees according to the pre-agreed pricing rules. The edge node will then return the task execution results to the terminal device and release the resources occupied by the tasks.

[0124] An Internet of Things MEC task offloading and allocation method according to an embodiment of the present disclosure, by proposing an offloading decision algorithm based on improved DQN, fully considering the load balancing of edge nodes and task delay constraints, realizes the collaborative computing between terminal devices and edge nodes, and maximizes the computing energy efficiency of the Internet of Things under the premise of meeting the delay limit; a cyclic combinatorial auction algorithm proposed in combination with economic theory is used to describe the resource allocation process in the edge computing scenario and optimize social welfare.

[0125] As Figure 2 shown, another embodiment of the present disclosure provides an MEC task offloading and resource allocation optimization system for the industrial Internet of Things. The Internet of Things includes terminal devices and edge nodes that are electrically connected to each other. The system includes:

[0126] An acquisition module 210, configured to acquire MEC tasks from the terminal devices;

[0127] A decision-making module 220, configured to determine the MEC task as a bidding candidate task or a local execution task according to a pre-established offloading decision algorithm; wherein, the offloading decision algorithm is pre-established by training a deep Q-network based on the computing energy efficiency of the Internet of Things;

[0128] An auction module 230, configured to determine the bidding candidate task as a winning bidding task or a re-decision task according to a pre-established auction algorithm; wherein, the auction algorithm is pre-established based on an optimal matching algorithm, a heuristic algorithm, and Vickrey auction rules;

[0129] An allocation module 240, configured to allocate the winning bidding task to a corresponding edge node for execution.

[0130] Exemplarily, as Figure 2 shown, the system further includes a building module 250, configured to build the offloading decision algorithm;

[0131] The building module 250 is specifically configured to:

[0132] Calculate the computing energy efficiency of the Internet of Things according to the computing latency and computing energy consumption of the terminal device and the edge node;

[0133] Define an optimization problem of the computing energy efficiency, and transform the optimization problem into a Markov model;

[0134] Train a deep Q-network in combination with the Markov model and the Epsilon-Greedy algorithm to obtain an offloading decision algorithm.

[0135] Exemplarily, the auction module 230 is specifically configured to:

[0136] Define a heuristic factor of the edge node;

[0137] Use the optimal matching algorithm to select the edge node with the highest heuristic factor, and determine the corresponding bidding candidate task as the winning bidding task;

[0138] Determine the payment price of the terminal device corresponding to the winning bidding task according to the Vickrey auction rules.

[0139] Specifically, an MEC task offloading and resource allocation optimization system for industrial Internet of Things in an embodiment of the present disclosure is used to implement the MEC task offloading and resource allocation optimization method for industrial Internet of Things described in the above embodiments. The specific implementation process has been described in detail in the above embodiments, and will not be elaborated herein.

[0140] An MEC task offloading and resource allocation optimization system for industrial Internet of Things according to an embodiment of the present disclosure proposes an offloading decision algorithm based on improved DQN, fully considering the load balance of edge nodes and task delay constraints, realizing collaborative computing between terminal devices and edge nodes, and maximizing the computing energy efficiency of the Internet of Things on the premise of meeting the delay limit; a cyclic combinatorial auction algorithm proposed in combination with economic theory is used to describe the resource allocation process in the edge computing scenario and optimize social welfare.

[0141] As Figure 3 shown, another embodiment of the present disclosure provides an electronic device, including:

[0142] At least one processor 301; and a memory 302 communicatively connected to the at least one processor 301 for storing one or more programs, which, when executed by the at least one processor 301, enable the at least one processor 301 to implement the above-mentioned MEC task offloading and resource allocation optimization method for industrial Internet of Things.

[0143] Wherein, the memory 302 and the processor 301 are connected by a bus. The bus may include any number of interconnected buses and bridges, and the bus connects various circuits of the one or more processors 301 and the memory 302 together. The bus may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, and thus will not be further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver may be one component or multiple components, such as multiple receivers and transmitters, providing units for communicating with various other devices on the transmission medium. The data processed by the processor 301 is transmitted over the wireless medium through the antenna. Further, the antenna also receives data and transmits the data to the processor 301.

[0144] The processor 301 is responsible for managing the bus and general processing, and may also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. The memory 302 may be used to store data used by the processor 301 when performing operations.

[0145] Another embodiment of the present disclosure provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned MEC task offloading and resource allocation optimization method for industrial Internet of Things.

[0146] Wherein, the computer-readable storage medium may be included in the system or electronic device of the present disclosure, or may exist alone.

[0147] A computer-readable storage medium can be any tangible medium that contains or stores a program. It can be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device. More specific examples include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, an optical fiber, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0148] A computer-readable storage medium can also include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Specific examples include, but are not limited to, electromagnetic signals, optical signals, or any suitable combination thereof.

[0149] It can be understood that the above embodiments are merely exemplary embodiments adopted to illustrate the principles of the present disclosure. However, the present disclosure is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present disclosure, and these modifications and improvements are also considered within the protection scope of the present disclosure.

Claims

1. An optimization method for MEC task offloading and resource allocation for industrial Internet of Things, the Internet of Things comprising terminal devices and edge nodes electrically connected to each other, characterized in that, The method includes: Obtaining an MEC task from the terminal device; Determining the MEC task as a bidding candidate task or a local execution task according to a pre-established offloading decision algorithm; wherein, the offloading decision algorithm is pre-established by training a deep Q-network based on the computing energy efficiency of the Internet of Things; Determining the bidding candidate task as a winning bidding task or a re-decision task according to a pre-established auction algorithm; wherein, the auction algorithm is pre-established based on an optimal matching algorithm, a heuristic algorithm, and Vickrey auction rules; Allocating the winning bidding task to a corresponding edge node for execution.

2. The method according to claim 1, characterized in that, The offloading decision algorithm is pre-established through the following steps: Calculating the computing energy efficiency of the Internet of Things according to the computing latency and computing energy consumption of the terminal device and the edge node; Defining an optimization problem of the computing energy efficiency and transforming the optimization problem into a Markov model; Training a deep Q-network in combination with the Markov model and the Epsilon-Greedy algorithm to obtain an offloading decision algorithm.

3. The method according to claim 2, characterized in that, The computing latency of the terminal device is: wherein is the offloading decision variable for task i k , is the computational amount of task i k , is the CPU resource ratio allocated by the terminal device to task i k , and f k is the local computing power of terminal device k; Task i k The computing energy consumption of the terminal device is: Wherein, is the computing power of the terminal device; The computing latency of the edge node j is: wherein, is the offloading decision variable for task i k , c ik is the computational complexity of task i k , is the proportion of computational resources allocated by edge node j to task i k , is the computing power of edge node j; The transmission delay of task i uploaded from the terminal device k to edge node j is as follows: Wherein, is the link length for task i k uploaded to edge node j, is the upload rate for task i k uploaded to edge node j; Upload task i from the terminal device k The transmission energy consumption to the edge node j is as follows: where p k is the transmission power of upload task i k to edge node j; Upload task i from the terminal device k The total execution delay to edge node j is as follows: The computing energy efficiency of the Internet of Things is: where B total is the total number of computing bits completed by the Internet of Things within the duration, and E total is the total energy consumption of the Internet of Things.

4. The method according to claim 3, characterized in that, The optimization problem is: maxU sys where the objective function is maxU sys i.e., to maximize the computing energy efficiency of the Internet of Things is the maximum tolerable delay of the task is the bandwidth ratio allocated by edge node j to task i k ​ 5. The method according to any one of claims 1 to 4, characterized in that The auction algorithm includes: Defining a heuristic factor of an edge node; Selecting, by using an optimal matching algorithm, the edge node with the highest heuristic factor and determining the corresponding bidding candidate task as a winning bidding task; Determining the payment price of the terminal device corresponding to the winning bidding task according to Vickrey auction rules.

Images