Edge data transmission optimization method based on energy consumption minimization and load balancing

Abstract

The invention discloses an edge data transmission optimization method based on energy consumption minimization and load balancing, which belongs to the technical field of edge computing, is realized through a system consisting of a central control server, an edge node, a transmission link and an SDN (Software Defined Network) controller, and comprises the following steps: S1, network topology modeling and data collection; s2, link selection optimization is carried out; s3, topological optimization and dynamic adjustment are carried out; s4, executing and controlling data transmission; s5, performing adaptive dynamic adjustment; and S6, the steps S1 to S5 are executed in a circulating mode. According to the method, a multi-target reinforcement learning algorithm is adopted, and joint optimization of energy consumption minimization and load balancing is achieved in edge data transmission; transmission links are dynamically adjusted and redundant nodes and links are closed, so that the overall energy consumption of the system is reduced; through link classification and load balancing optimization, the data transmission efficiency and the network stability are improved; and a heuristic factor and a Chebyshev scaling function are introduced, so that the convergence speed of the algorithm is accelerated.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to an edge data transmission optimization method based on energy consumption minimization and load balancing. Background Art

[0002] In the context of the Internet of Things (IoT), edge computing, as a distributed computing paradigm, aims to disperse computing tasks from traditional cloud computing centers to edge nodes close to data sources. In this way, edge computing can significantly reduce data transmission latency, improve real-time response capabilities, and reduce the load on central data centers.

[0003] In edge computing environments, the key challenge for data transmission is how to optimize network energy consumption and load distribution while meeting quality of service requirements. The complexity of this multi-objective optimization problem lies in the fact that energy consumption and load balancing are often conflicting: in order to achieve low energy consumption, it may be necessary to reduce the use of network devices, which may lead to excessive load on some links or nodes; in order to achieve load balancing, it may be necessary to activate more devices, which in turn increases energy consumption.

[0004] Although some studies have recognized the conflict between energy consumption and load balancing and have tried to combine them for optimization, most methods are still limited to using one of the objectives as a constraint, rather than true multi-objective optimization. For example, when performing load balancing, they use energy consumption minimization as a fixed constraint, but do not dynamically adjust it to adapt to the actual operation of the network. The result of this optimization strategy is that although load balancing can be achieved to a certain extent, in actual applications, the total energy consumption of the system is still difficult to achieve the optimal.

[0005] In addition, many existing optimization algorithms usually assume that the network topology and transmission requirements are static, that is, they do not change during the optimization process. However, in edge computing environments, network status and data transmission requirements are dynamically changing; IoT devices may have different data transmission requirements at different time periods, and network topology may also change due to device failures or the addition of new nodes. Static optimization algorithms are difficult to adapt to this dynamic environment, which may lead to poor optimization results in actual applications, and even fail to meet real-time requirements.

[0006] Therefore, how to provide a transmission optimization method with multi-objective joint optimization and dynamic adaptability is an urgent problem that technicians in this field need to solve. Summary of the invention

[0007] The purpose of the present invention is to provide an edge data transmission optimization method based on energy consumption minimization and load balancing to solve the problems in the background technology.

[0008] To achieve the above object, the present invention provides an edge data transmission optimization method based on energy consumption minimization and load balancing, comprising the following steps:

[0009] S1. Network topology modeling and data collection: The system performs comprehensive topology modeling and data collection on the network status;

[0010] S2. Link selection optimization: Design the state space and action space, obtain the Pareto approximate set through Pareto optimization, and use it as the global optimal solution. Construct a multi-objective reward function to classify each link selection strategy in the action space, introduce a heuristic factor to weight the result of the reward function, use the Chebyshev scalar function to comprehensively score and optimize multiple links, and select the optimal link;

[0011] S3, topology optimization and dynamic adjustment: Based on the results of S2, the system dynamically shuts down redundant nodes and links to further optimize the network topology. At the same time, the central control server monitors and adjusts the link status in real time through the SDN controller; specifically:

[0012] 1) Node shutdown: Nodes that do not require data transmission are placed in low-power mode until new data transmission requirements arise;

[0013] 2) Link optimization: close ports that are not involved in the current data transmission link and reduce power to save energy;

[0014] At the same time, the system has dynamic adjustment capabilities: the central control server monitors and adjusts the link status in real time through the SDN controller. When the load of a link is too high, the system will automatically select a new link and divert part of the traffic to other links;

[0015] S4, data transmission execution and control: After completing S3, the data transmission task is transmitted through the optimized link, and the SDN controller performs specific link configuration and bandwidth allocation to ensure the efficiency and stability of data transmission;

[0016] S5, Adaptive dynamic adjustment: The system automatically updates the Q value and Pareto approximation set according to the changes in the real-time network status and adjusts the transmission link;

[0017] S6. Execute S1 to S5 repeatedly.

[0018] Preferably, the specific steps of S1 are:

[0019] S11. Construct the minimum topology subset: The system analyzes the current network topology and determines the key links in the network that support the minimum data transmission. The key links are selected according to the following conditions:

[0020] 1) Link bandwidth utilization: Links with low bandwidth utilization are given priority to ensure data transmission needs without causing congestion;

[0021] 2) Node processing power and energy consumption: Only nodes with low computational load and low energy consumption are activated to avoid using edge nodes with high energy consumption;

[0022] S12, sending and receiving of detection data packets: the central control server periodically sends detection data packets to each edge node, and each detection data packet records the sending timestamp. After the edge node returns the response data packet, the central control server calculates the link delay based on the difference between the receiving time and the sending time;

[0023] S13, link bandwidth monitoring: use the central control server to count the amount of data transmitted on each link, and calculate the bandwidth utilization rate based on the transmission time; for example, if a link transmits X bytes of data within a certain period of time and takes Y seconds, then the bandwidth utilization rate of the link is X / Y;

[0024] S14. Node processing capacity and energy consumption: The central control server regularly monitors the CPU, memory and other resource usage of each edge node, collects the node's energy consumption data, and evaluates the load of each node to ensure that data transmission does not cause node overload.

[0025] Preferably, the specific steps of S2 are:

[0026] S21, using the current network topology and resource usage obtained in S1 as state information to construct a state space, each state includes: 1) the processing capacity and current load of each node; 2) the bandwidth utilization, transmission delay and energy consumption of each link; 3) the topology of the network;

[0027] S22, in each data transmission, the system will select one or more links to transmit data. The action of each state in the state space is to select a different transmission link. Each link selection strategy is used as an action to construct the action space;

[0028] S23. Combining multi-objective optimization and Q-learning methods, respectively construct energy consumption optimization reward function and load balancing reward function, then construct a multi-objective reward function with energy consumption optimization and load balancing as reward factors, and classify candidate links into four levels of high priority and low load, high priority and high load, low priority and low load, and give priority to high priority and low load links according to the multi-objective reward function;

[0029] The incentive factors are:

[0030] 1) Energy optimization: The lower the energy consumption of the link, the higher the reward value. The system gives priority to links with low energy consumption for data transmission, aiming to minimize the energy consumption of the overall network;

[0031] 2) Load balancing: The more balanced the load distribution of the link, the higher the reward value. By introducing the congestion factor of the link, the system can identify the links with lighter loads and give priority to these links for transmission;

[0032] S24, introducing a heuristic factor as a weighted proportional coefficient of the multi-objective reward function obtained in S23, and using the weighted result as the value of the final multi-objective reward function, further guiding the system to select a link in the global optimal solution and controlling the search for the solution space;

[0033] S25, action selection: For the link selected in S24, a link selection mechanism based on Chebyshev scalar function is used to select the link with the minimum energy consumption and the best load balancing after multiple iterations.

[0034] Preferably, S21 to S25 constitute a multi-objective reinforcement learning algorithm.

[0035] Preferably, in S23, the specific steps of the multi-objective optimization method are:

[0036] Considering both energy consumption minimization and load balancing, the Pareto approximate set is obtained through Pareto optimization as the global optimal solution. In the actual optimization process, it is very difficult to find the exact optimal set, so the Pareto approximate set is used to represent the set close to the global optimal solution.

[0037] Assume that the network topology is represented by a directed graph G, G = (V, E, C), where V is the node set, E is the link set, and C is the link capacity set;

[0038] Define the congestion factor θ e Link e ij The flow rate on ij With capacity c ij The ratio is expressed as:

[0039]

[0040] Where d∈D represents any destination d in the transmission destination set D. Indicates that the transmission traffic with d as the destination is on link e ij The sum of the above;

[0041] The two main goals of data transmission optimization are to minimize the total energy consumption P total And to achieve load balancing, we get the maximum congestion factor θ on all links. eThe minimum value of is:

[0042] 1) Minimize the total energy consumption P total , expressed as:

[0043]

[0044] In the formula, y i Represents node v i Is it active? Represents node v i Energy consumption, x ij Indicates link e ij Is it active? and Respectively represent the node v i and v j Link energy consumption;

[0045] 2) Minimize the maximum congestion factor θ e , expressed as:

[0046]

[0047] Where E is the set of links in the current network topology, and e is any link in E;

[0048] 3) When optimizing, the following constraints must also be met:

[0049] Flow conservation constraint,The flow conservation constraint of any node i is expressed as:

[0050]

[0051] Where, t d represents the total bandwidth requirement of destination d, and DEM is the set of node pairs with transmission requirements in the current network topology, such as<i,d> This means that there is a transmission demand from node i to destination node d.

[0052] Link capacity constraint: the total flow of each link cannot exceed its capacity, expressed as:

[0053]

[0054] Node activation constraint: A node can be active only when all its inbound and outbound links are activated, expressed as:

[0055]

[0056] Where M is a large number that ensures that a node is active only when all incoming and outgoing links are activated.

[0057] Preferably, in S23, the energy consumption optimization reward function is:

[0058]

[0059] The load balancing reward function is:

[0060]

[0061] In the formula, l represents the link, res(l) is the remaining bandwidth of link l, dem(i) is the bandwidth requirement of transmission demand i, PH is the set of candidate links, β, β1 and β2 are the proportional coefficients for adjusting different target weights, and r is the bandwidth surplus rate of the link. The calculation formula is: total(l) is the total bandwidth of the link, θ is the bandwidth utilization of the link, and the calculation formula is θ avg is the average bandwidth utilization of all selected links;

[0062] The multi-objective reward function is:

[0063]

[0064] The four conditional parts in the multi-objective reward function, the condition "res(l) <d emi " corresponds to the link that does not meet the conditions, and the reward is 0; the condition "res(l)≥d emi ,l∈PH” corresponds to high priority low load and high priority high load links, and the specific reward is determined by the load r; the condition Rewards for low-priority, low-load links; Conditions Rewards corresponding to low priority high load links.

[0065] Preferably, in S24, the heuristic factor is defined as:

[0066]

[0067] Where η l,cur is a variable with a value of 0 or 1. When it is 1, it means that the link l selected this time is in the currently known optimal solution. When it is 0, it means that the link l selected this time is not in the currently known optimal solution. n is the current number of iterations, and N is the maximum number of iterations. Used to represent the total energy consumption of the current global optimal solution, PT gb represents the link set of the current global optimal solution, θ e is the link congestion factor;

[0068] η l,curThe weight of the current global optimal solution is normalized so that the link in the current global optimal solution obtains more rewards through the reward function, guiding the model to select the link in the current optimal solution;

[0069] At the beginning of the iteration, the value of n / N is very small, and the value of h is also small, which has little impact on the reward function, and can obtain a wider search range, which is more conducive to exploration. As n / N gradually increases, the value of h also gradually increases, and the multi-objective reinforcement learning algorithm selection will focus more on the global optimal solution set, which helps to improve the convergence speed of the algorithm;

[0070] In the initial stage of the algorithm, the global optimal solution is defined as an empty set. In the initial iteration, the algorithm generates solutions and calculates the multi-objective performance values ​​of these solutions, and then screens out non-dominated solutions (that is, solutions that are not surpassed by other solutions in all optimization objectives) and adds them to the global optimal solution. After each round of iteration, the global optimal solution is updated by adding new non-dominated solutions and removing dominated solutions (surpassed in some optimization objectives). This process ensures that the global optimal solution is always the current optimal set of non-dominated solutions, and gradually approaches the final Pareto optimal solution as the number of iterations increases.

[0071] Preferably, the specific steps of S25 are as follows:

[0072] 1) Generate multiple candidate links for the link selected in S24 according to the demand, score each link according to the energy consumption and load balancing of the link, and calculate the Q value for decision-making for the energy consumption optimization reward function and the load balancing reward function in S23;

[0073] 2) Use the Chebyshev scalar function to optimize the comprehensive score of each link and select the link with the highest total score. The scalar formula is:

[0074]

[0075] In the formula, is the scalarized Q value, ω i is the weight of the ith target, Represents a state-action pair (s t ,a t ) is the Q value on the i-th target, z i is the ideal value of the ith target, Indicates that the maximum value of the deviation of all targets is sought;

[0076] The Chebyshev scalarization method is used to calculate the weighted deviation of each target. If the deviation of a target is particularly large, it will dominate the final scalar value, driving the optimization algorithm to prioritize improving this target to reduce the deviation. This mechanism ensures the balance between all targets and avoids the large deviation of some targets being ignored.

[0077] 3) According to the scalarized Value, selected in the current state The action with the smallest value a t As the action with the minimum energy consumption and the best load balance, the next transmission is performed.

[0078] Preferably, the specific implementation process of S4 is:

[0079] 1) Route adjustment: The SDN controller transmits data packets through optimized links according to the instructions of the central control server. If a link is congested, the system will immediately recalculate and allocate a new link to avoid transmission bottlenecks;

[0080] 2) Bandwidth allocation: The SDN controller adjusts bandwidth allocation according to the real-time status of the link to ensure that high-load links are not overly congested. The system adjusts the transmission bandwidth of each link in real time to maintain the overall stability of the network.

[0081] Preferably, the specific implementation process of S5 is:

[0082] 1) Real-time monitoring: The central control server continuously monitors the bandwidth, latency, and load of nodes and links;

[0083] 2) Dynamic adjustment: When it is detected that the link utilization exceeds the threshold, the system will automatically re-plan the link to ensure that the data is correctly scheduled.

[0084] Therefore, the edge data transmission optimization method based on energy consumption minimization and load balancing of the present invention has the following beneficial effects:

[0085] (1) The present invention takes energy consumption and load balancing as optimization goals, and ensures that these two goals can be optimized simultaneously during data transmission by designing a reasonable reward function. The present invention gradually approaches the optimal solution by multi-step state updates, and reduces the number of occupied devices by constructing a minimum topology subset, thereby minimizing energy consumption. In addition, during the load distribution optimization process, the load on the selected link is balanced by minimizing the maximum congestion factor to avoid overloading of certain links or nodes.

[0086] (2) The present invention adopts a multi-objective optimization algorithm based on reinforcement learning, comprehensively considers the two objectives of energy consumption and load balancing when selecting transmission links, and selects the transmission link with the least occupied devices and links by dynamically updating the Pareto approximation set, thereby minimizing the number of active devices and their power consumption. In practical applications, this means that the computing resources and transmission resources in the network are more reasonably allocated, thereby greatly reducing the power consumption during the transmission process. This optimization effect is particularly obvious in large-scale data transmission scenarios, enabling the system to maintain low energy consumption while meeting the needs of efficient data transmission.

[0087] (3) The present invention also proposes a link classification mechanism that divides candidate links into four different levels so that the optimal link can be selected during the link construction process, which helps to optimize link selection and reduce transmission delay and energy consumption while ensuring load balancing. The present invention also effectively solves the problem of uneven distribution of data traffic on each link in traditional networks through a load balancing optimization mechanism; the introduction of multi-objective optimization ideas not only optimizes transmission links, but also achieves a more balanced distribution of traffic in the network. By classifying and real-time monitoring of links, load balancing of each link is ensured, the frequency of use of overloaded links is reduced, and transmission delay and data congestion are significantly reduced. This optimization strategy greatly improves the overall throughput of the network, enabling the system to complete more data transmission tasks in a shorter time, thereby improving transmission efficiency and service quality.

[0088] (4) In order to speed up the convergence of the algorithm, the present invention introduces a heuristic factor to control the search of the solution space, and by strengthening the guiding effect of the existing optimal solution, the optimization effect of the algorithm in a complex network environment is improved; it can find a transmission link close to the global optimal in a short time, avoiding the problem of local optimal solution, and significantly shortening the calculation and decision-making time. This feature enables the present method to quickly adapt to changes in the network environment, especially in data transmission scenarios that require real-time optimization, and can ensure the efficient operation of the system.

[0089] (5) In order to effectively select weights in multi-objective optimization, the present invention adopts a Chebyshev scalarization function based on roulette wheel selection to optimize the weight selection process. This strategy can make reasonable trade-offs between multiple objectives, ensuring that energy consumption and load balancing can be taken into account during the optimization process, thereby obtaining a higher quality solution.

[0090] (6) The present invention uses the Q-learning algorithm in reinforcement learning to dynamically adjust the transmission link and load distribution according to the actual operation status and transmission requirements of the network. When certain links or devices fail, the system can respond quickly and reallocate the transmission link to ensure the continuity and stability of data transmission. In addition, the load balancing optimization mechanism of the present invention enables the system to effectively cope with the impact of burst traffic and avoid network crashes caused by overload of individual links. This adaptability enables the system to maintain an efficient and stable operation state in the face of complex network environments.

[0091] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 It is a general process framework diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0093] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0094] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0095] The method protected by the present invention is based on the optimization system. The system includes a central control server, edge nodes, transmission links and SDN controllers. The central control server is the core control unit of the entire system. As the core control unit of the system, the central control server is responsible for real-time monitoring of all edge nodes and links in the entire network. The central server collects key parameters of each node in real time by regularly sending detection data packets. The central control server not only needs to collect data from each edge node regularly, but also needs to dynamically adjust the path and network topology based on these data to ensure that energy consumption can be minimized and load can be evenly distributed during data transmission.

[0096] Edge nodes are located near data sources and are responsible for preprocessing data and communicating with central servers and other nodes. Transmission links connect edge nodes and data centers or other edge nodes, forming the physical basis of the entire network. During each data transmission, the system needs to select the optimal transmission path to minimize the energy consumption of nodes and links while ensuring that the load is evenly distributed throughout the network.

[0097] The SDN controller is used to execute the path adjustment instructions issued by the central control server. The SDN controller controls the status of each network device through the OpenFlow protocol and flexibly adjusts the traffic scheduling, bandwidth allocation and topology configuration of the network device. The centralized control mechanism of SDN makes network adjustment more flexible and efficient.

[0098] The following uses the above system and an edge data transmission optimization method based on energy consumption minimization and load balancing proposed by the present invention to illustrate its important role in different usage scenarios.

[0099] Example 1

[0100] Apply the above system and optimization method in smart factories:

[0101] In a smart factory, multiple edge nodes are deployed on different production equipment. These nodes collect real-time equipment status information such as temperature, vibration, and operating status through sensors. The central control server is responsible for monitoring the processing capacity, bandwidth utilization, and energy consumption of each node. The transmission link connects each node to the central server, and the SDN controller is used to control the traffic scheduling and link selection of each node.

[0102] After the system was fully deployed, the method protected by the present invention was used for testing, and the results showed that data transmission was efficient, energy consumption was low, network stability was good, and system latency was low.

[0103] This means that when it is applied to smart factories, during data transmission, the system adjusts the transmission link according to the real-time status, reduces congestion, and ensures efficient transmission of production data. It reduces the energy consumption of smart factory equipment while improving network stability and effectively reduces the overall delay of the system.

[0104] Example 2

[0105] Apply the above system and optimization method in remote monitoring system for efficient data transmission:

[0106] In the remote monitoring system, multiple cameras are used as edge nodes to collect environmental video data in real time. The central control server is deployed in the data center and is responsible for receiving and processing data from all edge nodes. The transmission link in the system connects the monitoring camera and the central control server, and the SDN controller controls the flow and priority of the data.

[0107] The method protected by the present invention is used for testing, and the results show that data transmission is stable, video monitoring is clear and real-time, and energy consumption is low.

[0108] This shows that when it is applied to remote monitoring systems, the system gives priority to low-power links and avoids high-load links during data transmission, reducing energy consumption. The load balancing mechanism in the optimization method improves the stability of data transmission and ensures the real-time, high efficiency and clarity of video traffic. Through the calculation of the system, the transmission links are reasonably allocated, which saves power resources and is suitable for remote monitoring scenarios with limited power supply.

[0109] Example 3

[0110] Apply the above system and optimization method to intelligent traffic management for link optimization:

[0111] In the intelligent transportation system, traffic lights and monitoring equipment constitute edge nodes, and sensors collect real-time traffic flow data. The central control server is responsible for analyzing and processing this data to provide support for traffic scheduling. The transmission link connects each traffic node with the central server, and the SDN controller is used to control traffic scheduling.

[0112] The method protected by the present invention is used for testing, and the results show that data transmission is smooth and time-effective during peak hours.

[0113] This means that when it is applied to intelligent traffic management, the optimization method can monitor the data flow distribution in the network in real time, ensuring that the selection of transmission links not only meets the requirements of low energy consumption but also maintains load balance, ensuring that data flows can be transmitted smoothly during peak hours and that there will be no signal delays due to network congestion.

[0114] Therefore, the present invention adopts the above-mentioned edge data transmission optimization method based on energy consumption minimization and load balancing, and adopts a multi-objective reinforcement learning algorithm to realize the joint optimization of energy consumption minimization and load balancing in edge data transmission; reduces the overall energy consumption of the system by dynamically adjusting the transmission link and shutting down redundant nodes and links; improves data transmission efficiency and network stability through link classification and load balancing optimization; introduces heuristic factors and Chebyshev scalarization functions to accelerate the convergence speed of the algorithm.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An edge data transmission optimization method based on energy consumption minimization and load balancing is implemented by a system consisting of a central control server, edge nodes, transmission links and an SDN controller, characterized in that: The following steps are involved: S1. Network topology modeling and data collection: The system performs comprehensive topology modeling and data collection on the network status; S2. Link selection optimization: Design the state space and action space, obtain the Pareto approximate set through Pareto optimization, and use it as the global optimal solution. Construct a multi-objective reward function to classify each link selection strategy in the action space, introduce a heuristic factor to weight the result of the reward function, use the Chebyshev scalar function to comprehensively score and optimize multiple links, and select the optimal link; S3, topology optimization and dynamic adjustment: Based on the results of S2, the system dynamically shuts down redundant nodes and links to further optimize the network topology. At the same time, the central control server monitors and adjusts the link status in real time through the SDN controller; S4, data transmission execution and control: After completing S3, the data transmission task is transmitted through the optimized link, and the SDN controller performs specific link configuration and bandwidth allocation to ensure the efficiency and stability of data transmission; S5, Adaptive dynamic adjustment: The system automatically updates the Q value and Pareto approximation set according to the changes in the real-time network status and adjusts the transmission link; S6. Execute S1 to S5 repeatedly.

2. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 1, characterized in that: The specific steps of S1 are: S11. Constructing the minimum topology subset: The system analyzes the current network topology and determines the key links in the network that support the minimum data transmission; S12, sending and receiving detection data packets: the central control server periodically sends detection data packets to each edge node, and after the edge node returns a response data packet, the link delay is calculated; S13, link bandwidth monitoring: using the central control server to calculate the bandwidth utilization of each link; S14. Node processing capacity and energy consumption: The central control server regularly monitors the resource usage of each edge node, collects the node's energy consumption data, and evaluates the load of each node to ensure that data transmission does not cause node overload.

3. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 1, characterized in that: The specific steps of S2 are: S21, using the current network topology and resource usage obtained in S1 as state information to construct a state space, each state includes: 1) the processing capacity and current load of each node; 2) the bandwidth utilization, transmission delay and energy consumption of each link; 3) the topology of the network; S22, the action of each state in the state space is to select a different transmission link, and each link selection strategy is used as an action to construct the action space; S23. Combining multi-objective optimization and Q-learning methods, respectively construct energy consumption optimization reward function and load balancing reward function, then construct a multi-objective reward function with energy consumption optimization and load balancing as reward factors, and classify candidate links into four levels of high priority and low load, high priority and high load, low priority and low load, and give priority to high priority and low load links according to the multi-objective reward function; S24, introducing a heuristic factor as a weighted proportional coefficient of the multi-objective reward function obtained in S23, and using the weighted result as the value of the final multi-objective reward function to further guide the system to select a link in the global optimal solution; S25, action selection: For the link selected in S24, a link selection mechanism based on Chebyshev scalar function is used to select the link with the minimum energy consumption and the best load balancing after multiple iterations.

4. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 3 is characterized in that: In S23, the energy consumption optimization reward function is: The load balancing reward function is: In the formula, l represents the link, res(l) is the remaining bandwidth of link l, dem(i) is the bandwidth requirement of transmission demand i, PH is the set of candidate links, β, β1 and β2 are the proportional coefficients for adjusting different target weights, and r is the bandwidth surplus rate of the link. The calculation formula is: total(l) is the total bandwidth of the link, θ is the bandwidth utilization of the link, and the calculation formula is θ avg is the average bandwidth utilization of all selected links; The multi-objective reward function is:

5. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 4, characterized in that: In S24, the heuristic factor is defined as: Where η l,cur is a variable, n is the current number of iterations, N is the maximum number of iterations, Used to represent the total energy consumption of the current global optimal solution, PT gb represents the link set of the current global optimal solution, θ e is the link congestion factor.

6. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 5, characterized in that: The specific steps of S25 are as follows: 1) Generate multiple candidate links for the link selected in S24 according to the demand, score each link according to the energy consumption and load balancing of the link, and use the Q-learning method to calculate the Q value for decision-making based on the energy consumption optimization reward function and load balancing reward function in S23; 2) Use the Chebyshev scalar function to optimize the comprehensive score of each link and select the link with the highest total score. The scalar formula is: In the formula, is the scalarized Q value, ω i is the weight of the ith target, Represents a state-action pair (s t ,a t ) is the Q value on the i-th target, z i is the ideal value of the ith target, Indicates that the maximum value of the deviation of all targets is sought; 3) According to the scalarized Value, selected in the current state The action with the smallest value a t As the action with the minimum energy consumption and the best load balance, the next transmission is performed.

7. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 6, characterized in that: In step 1) of S25, the Q value calculation formula is: In the formula, s t and a t Represent the current state and action respectively, α t is the learning rate, r t is the current instant reward, γ t is the discount factor, s t′ and a t′ They are the next state and action after taking the action.

8. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 1, characterized in that: The specific steps of S4 are: 1) Route adjustment: The SDN controller transmits the data packet through the link selected by S25 according to the instructions of the central control server. If a link is congested, the system will immediately recalculate and allocate a new link to avoid transmission bottlenecks; 2) Bandwidth allocation: The SDN controller adjusts bandwidth allocation according to the real-time status of the link to ensure that high-load links are not overly congested. The system adjusts the transmission bandwidth of each link in real time to maintain the overall stability of the network.

9. The edge data transmission optimization method based on energy consumption minimization and load balancing according to claim 1, characterized in that: The specific steps of S5 are: 1) Real-time monitoring: The central control server continuously monitors the bandwidth, latency, and load of nodes and links; 2) Dynamic adjustment: When it is detected that the link utilization exceeds the threshold, the system automatically replans the link to ensure that the data is correctly scheduled.

Images