Self-adaptive network flow optimization method and system based on edge calculation

Abstract

The invention relates to the technical field of network flow optimization, in particular to a self-adaptive network flow optimization method and system based on edge computing, and the method comprises the following steps: 1, building a flow relation model to analyze the states of a data flow in a storage network and a computing network; step 2, seamless connection of a network hardware interface is realized; 3, constructing a topological structure mapping table, and dynamically managing and analyzing by using storage resources to provide a basis for optimal scheduling; step 4, formulating an optimal scheduling strategy according to flow prediction and positioning results; step 5, optimizing the data rate and the packet size by using a traffic warping method to improve the data transmission efficiency; 6, deploying a flow marking and priority mechanism, and guaranteeing the low-delay flow; and step 7, heterogeneous equipment collaborative scheduling optimization based on reinforcement learning is realized, and intelligent and efficient management of network traffic is realized by transferring calculation and decision-making capabilities to the edge of the network and combining a self-adaptive traffic management strategy and a reinforcement learning technology.

Description

Technical Field

[0001] The present invention relates to the technical field of network traffic optimization, and in particular to an edge computing-based adaptive network traffic optimization method and system thereof. Background Art

[0002] With the rapid development of the Internet of Things and 5G technology, network traffic has exploded, and traditional network architecture is facing unprecedented challenges. Emerging applications such as smart cities, industrial Internet, and Internet of Vehicles have put forward higher requirements for the real-time, reliability, and efficiency of the network. In this context, the optimization and management of network traffic has become a key issue that needs to be solved urgently.

[0003] Currently, the closest existing technologies mainly rely on centralized cloud computing architecture. This approach does excel in large-scale data processing, but its inherent flaws become increasingly prominent in the face of increasingly complex network environments. First, the centralized architecture suffers from serious network latency issues. When a large amount of data needs to be transmitted to a remote data center for processing, significant transmission delays are inevitable. Second, traditional methods have difficulty coping with dynamic changes in the network environment. Fixed traffic management strategies cannot adapt to fluctuations in network conditions, resulting in inefficient resource utilization. In addition, with the increase in the number of edge devices, a unified traffic management strategy is difficult to meet the diverse needs of different types of devices. Finally, traditional methods perform poorly in terms of energy efficiency, and the remote transmission and centralized processing of large amounts of data leads to high energy consumption. Summary of the invention

[0004] In view of the above problems, the present invention proposes an adaptive network traffic optimization method and system based on edge computing. The method aims to solve technical problems such as high network latency, low resource utilization efficiency, difficulty in adapting to dynamic network environments, and high energy consumption. By delegating computing and decision-making capabilities to the edge of the network, combined with adaptive traffic management strategies and reinforcement learning technology, intelligent and efficient management of network traffic is achieved.

[0005] The present invention provides an adaptive network traffic optimization method based on edge computing, comprising the following steps:

[0006] Step 1. Establish a traffic relationship model to analyze the status of data flow in the storage network and computing network; Step 2. Realize seamless connection of network hardware interfaces, deploy adaptive bridging methods, and complete network communication; Step 3. Construct a topology mapping table, and use dynamic management and analysis of storage resources to provide a basis for optimized scheduling; Step 4. Formulate traffic scheduling rules based on reinforcement learning, and formulate the best scheduling strategy based on traffic prediction and positioning results; Step 5. Use traffic regularization methods to optimize data rate and packet size to improve data transmission efficiency; Step 6. Deploy traffic marking and priority mechanisms to ensure low-latency flows; Step 7. Implement collaborative scheduling optimization of heterogeneous devices based on reinforcement learning.

[0007] Preferably, the detailed steps of step 1 are as follows:

[0008] 1-1. Based on the timestamp and address information of the data packet, a data traffic relationship model in the storage network and computing network is established, including:

[0009] a. For the storage network part, establish the equation of storage network storage capacity changing over time. The storage capacity change rate function is recorded as H. According to the storage capacity change rate function calculation formula, the storage network storage capacity changing over time can be expressed as:

[0010]

[0011] H(t)=β·e -λt ·B·V

[0012] Where t represents any time, S(t) represents the storage capacity at time t, S0 represents the initial storage capacity, H(τ) represents the storage capacity change rate function, β and λ represent storage change rate parameters, B represents the data block size, and V represents the average data write rate;

[0013] b. For the calculation network part, establish an expression for calculating the change of network throughput over time. The average link utilization is:

[0014]

[0015] Where L represents the length of each packet, T represents the throughput, B represents the bandwidth capacity, and N represents the number of links on each path;

[0016] The relationship between the throughput and utilization of each path is:

[0017]

[0018] According to the above steps, the relationship function between data transmission rate and computing time and storage time can be expressed as:

[0019]

[0020] Wherein, R(t) represents the data transmission rate at time t, α, β, γ represent the parameters of the throughput change rate, and V represents the average data writing rate.

[0021] Preferably, the detailed steps of step 2 are as follows:

[0022] 2-1. Define the timestamp of the data packet. Assume that the timestamp of the data packet is T1, T2, T3, ..., S n , the transmission rate is recorded as A, the size of the data flow in the network is recorded as B, and the expected value of the data flow length in the network is:

[0023]

[0024] Among them, T1, T2, T3, ...S n It is represented as the timestamp of the data packet, B is the length of each packet, X is the length of the data stream, and E(X) is the expected value of the length of the data stream;

[0025] 2-2. Based on the adaptive bridging method, the seamless connection of the network hardware interface is completed, and the steps include:

[0026] a. Define the total number of transmission nodes as K, and the edge between nodes i and j as B i，j is the link bandwidth capacity connecting nodes i and j, denoted as B i，j =w ij , the throughput of node i is denoted as T i , its traffic marking weight value is: Among them, W i represents the traffic marking weight value of node i, T i represents the throughput of node i, K represents the total number of transmission nodes;

[0027] b. Determine the congestion of the current network traffic at each node: compare the traffic size with the current network status. If the current network traffic value is less than or equal to the current network status, no data packets will be sent; if the current network traffic value is greater than the current network status, node i will select the current traffic value less than the network status B. i，j The node of is taken as the next hop node, and the next node of node j is calculated:

[0028] Next(j)=argmax k∈N(j) (B j，k -F j，k )

[0029] Among them, Next(j) represents the next node of the node, N(j) represents the set of neighbor nodes of the node, and B j，krepresents the link bandwidth capacity from node to node k, F j，k Represents the current flow from node j to node k;

[0030] If the node is the destination node, the data reaches the destination node through the data link, and the data packet is passed to the storage network resource for storage or the computing network resource for calculation; if the node is not the destination node, the data packet will be sent to the next hop node for the next iteration; if the next hop node does not exist, it returns to node i and broadcasts a data retransmission message to supplement the link load capacity;

[0031] 2-3. Select the best bridge point to complete seamless network connection.

[0032] Preferably, the detailed steps of step 3 are as follows:

[0033] 3-1. Define the input flow of computing network node k as I k , the output flow is O k , then the flow of node k in the computing network can be calculated as:

[0034]

[0035] Among them, k represents the node, F k represents the traffic of node k, I k represents the input flow of node k, O k represents the output flow of node k, N(k) represents the set of neighbor nodes of node k, and F m，k represents the traffic from the previous hop node m to node k, F k，n It represents the traffic that enters node k and reaches the next hop n of node k within a period of time;

[0036] 3-2. When the traffic between nodes in the computing network meets the transmission rules, the computing network nodes and the storage network nodes are connected, and a topology mapping table is established.

[0037] Preferably, the detailed steps of step 4 are as follows:

[0038] 4-1. In the storage network, since computing resources need to be consumed, a threshold function is set, which is expressed as:

[0039]

[0040] Among them, α, β represent the parameters of the threshold function, x represents the input data, Represents the output value of the threshold function;

[0041] 4-2. Within the time threshold, analyze the dynamic changes of network resources, and based on the predicted value of the traffic relationship model, obtain the bottleneck of the storage network and realize the traffic distribution.

[0042] Preferably, the detailed steps of step 5 are as follows:

[0043] 5-1. Weighting the data rate and packet size, the transmission time is expressed as:

[0044]

[0045] Where X and Y represent the total data rate and total packet size of the transmission, T represents the transmission time, M represents the packet size, and N represents the number of links in the network;

[0046] The transmission rate is calculated using an exponential weighted average algorithm, and the average exponential weight T is used as the basis for adjusting the ratio. The formula can be expressed as:

[0047] T(t)=β·T(t-1)+(1-β)·V(t)

[0048] Where β represents the weight coefficient, T(t) represents the reference of the adjustment ratio at time t, and V(t) represents the current data rate at time t;

[0049] 5-2. According to the average index weight T calculated in 5-1 and the dynamically adjusted data transmission speed rate V, calculate the average value W of the adjusted data transmission rate. The formula is as follows:

[0050] W=β·T+(1-β)·V

[0051] Wherein, W represents the average value of the adjusted data transmission rate, V represents the current data rate, T represents the adjustment ratio, and β represents the weight coefficient.

[0052] Preferably, the detailed steps of step 6 are as follows:

[0053] 6-1. Add a priority tag field to the header of the transmitted frame and assign different priority tags to the data streams;

[0054] 6-2. The scheduler calculates the remaining time (Time To Go, TTG) of each data flow in the queue. The formula is as follows:

[0055]

[0056] Among them, TTG represents the remaining time of each data flow in the queue, L d Indicates the target node of the data flow, L c Indicates the current node, O k represents the output flow of node k, L represents the length of data flow, and M represents the length of each packet;

[0057] 6-3. When sending data streams to queue, they are sent in the order of the queue according to the data stream priority tag.

[0058] Preferably, the detailed steps of step 7 are as follows:

[0059] 7-1. Model heterogeneous devices and match them with storage networks and computing networks; at the same time, assign system parameters to heterogeneous devices; the system parameters include: memory, CPU speed, bandwidth, power consumption, storage resources and data flow;

[0060] 7-2. Each network traffic node of the network device is trained through heterogeneous devices to obtain traffic scheduling results;

[0061] 7-3. Set the reward function to achieve optimal strategy learning: the reward function includes a network delay reward function T based on the storage device reward and a reward function C based on the network delay of the computing device reward for:

[0062] T reward =-W s ·T s

[0063] C reward =-W c ·T c

[0064] Among them, T s and T c represents the storage network and computing network delay, W s and W c Expressed as an optimizable parameter; select the corresponding reward function through the system parameters, distribute the traffic to the storage network and the computing network, and use the optimized parameter W as the benchmark to obtain the maximum reward function Q i :

[0065]

[0066] Among them, R k represents the benefit of network k, γ represents the discount factor, s′ represents the next state, a represents the action, K represents the network set, A represents the action set, and Q(s′, a) represents the state-action value function;

[0067] 7-4. Use the Q table in reinforcement learning to iterate until the maximum number of iterations is reached to obtain a trained network device;

[0068] 7-5. Trained network devices schedule computing tasks and storage tasks to edge devices.

[0069] An adaptive network traffic optimization system based on edge computing includes a storage network, a computing network and an edge device, wherein the edge device includes a traffic shaping module, a traffic marking module and a traffic scheduling module. The storage network includes a storage network device, and the computing network includes a computing network device. The edge device is connected to the storage network device and the computing network device through a wired connection or a wireless connection, respectively, and the storage network device and the computing network device are connected through a wired connection or a wireless connection. The traffic between the edge device and the storage network device and the computing network device is transmitted after being processed by the traffic shaping module, the traffic marking module and the traffic scheduling module.

[0070] Preferably, the storage network device is an intelligent device based on cloud computing and distributed storage methods, which is accessed in the form of data streams, and flexible scheduling of computing tasks and storage tasks is achieved through storage resource management; the computing network device is an intelligent device based on machine learning methods, which is accessed in the form of data streams, and flexible scheduling of computing tasks and storage tasks is achieved through computing resource management; the edge device is an intelligent device based on edge computing methods, which is accessed in the form of data streams, and load balancing of low-latency data is achieved by running reinforcement learning and traffic management.

[0071] Advantages and beneficial effects of the present invention:

[0072] First, the present invention realizes local data processing by deploying traffic regularization modules, traffic marking modules and traffic scheduling modules on edge devices. This distributed architecture greatly reduces the delay of data transmission. For example, in the intelligent transportation system, the control decision of traffic lights can be made directly on the edge device at the intersection without transmitting data to the remote data center, thus achieving a millisecond-level response speed.

[0073] Secondly, the flow relationship model analysis and topology structure mapping table construction of the present invention lay the foundation for the efficient use of network resources. By analyzing the status of the storage network and the computing network in real time, the system can accurately grasp the distribution and usage of network resources. Based on this information, the flow scheduling module can formulate the optimal resource allocation strategy. For example, in the industrial Internet of Things scenario, the system can dynamically adjust the data processing priority and resource allocation according to the real-time status of the production line, ensuring that key production data is processed in a timely manner while avoiding waste of resources.

[0074] Furthermore, the present invention introduces traffic scheduling rules based on reinforcement learning, which enables the system to continuously learn and adapt to complex and changing network environments. The reinforcement learning algorithm gradually forms a highly adaptable scheduling strategy through continuous trial and optimization. The advantage of this method is that it does not need to predefine all possible network conditions, but can learn and improve autonomously. For example, when faced with sudden network traffic peaks, the system can quickly adjust resource allocation to ensure the stable operation of the network.

[0075] In addition, the flow marking and priority mechanism of the present invention ensures that different types of data are processed differently. By assigning priorities to different data flows, the system can prioritize the operation of key services when resources are limited. This is particularly important in complex smart city management systems, for example, to ensure that the processing of emergency events will not be affected by ordinary data flows.

[0076] Finally, by comprehensively applying the above technologies, the present invention significantly improves the energy efficiency of the entire system. Edge computing reduces a large amount of unnecessary data transmission, while intelligent resource scheduling ensures the efficient use of computing resources. This not only reduces energy consumption, but also reduces carbon emissions, which meets the requirements of sustainable development.

[0077] In general, the present invention constructs an efficient, flexible and intelligent network traffic optimization system through the organic combination of advanced technologies such as edge computing, adaptive traffic management and reinforcement learning. This system not only solves the urgent problems facing current network management, but also provides strong technical support for the development of the future intelligent and networked world. Its wide application will promote the rapid development of smart cities, industrial Internet, Internet of Vehicles and other fields, and bring a more efficient, smarter and more sustainable future to society. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a logic block diagram of the system of the present invention.

[0079] Figure 2 It is a logic block diagram of the method flow of the present invention.

[0080] Figure 3 A logical block diagram of the steps for establishing a flow relationship model for the present invention.

[0081] Figure 4 The present invention is a logic block diagram of the steps for realizing seamless connection of network hardware interface.

[0082] Figure 5 This is a logic block diagram of the steps of constructing a topology structure mapping table of the present invention.

[0083] Figure 6 A logic block diagram of the steps for formulating traffic scheduling rules based on reinforcement learning according to the present invention.

[0084] Figure 7 This is a logic block diagram of the steps of the traffic regularization method used in the present invention.

[0085] Figure 8 This is a logic block diagram of the steps for deploying traffic marking and priority mechanism of the present invention.

[0086] Fig. 9 This is a logic block diagram of the optimization steps for implementing the collaborative scheduling of heterogeneous devices based on reinforcement learning according to the present invention. DETAILED DESCRIPTION

[0087] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0088] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0089] See also Figure 1-9 The present invention relates to an adaptive network traffic optimization method based on edge computing and a system thereof. With the rapid development of the Internet of Things and 5G technology, network traffic has shown explosive growth, and the traditional centralized network architecture has been unable to meet the growing data processing needs. In order to solve this problem, the present invention proposes an adaptive network traffic optimization method based on edge computing and a system thereof, aiming to improve network efficiency, reduce latency, and achieve smarter resource allocation.

[0090] First, let us elaborate on the adaptive network traffic optimization method based on edge computing, which includes the following steps: Step 1. Establish a traffic relationship model to analyze the status of data flow in the storage network and computing network; Step 2. Realize seamless connection of network hardware interfaces, deploy adaptive bridging methods, and complete network communication; Step 3. Construct a topology mapping table, and use dynamic management and analysis of storage resources to provide a basis for optimized scheduling; Step 4. Formulate traffic scheduling rules based on reinforcement learning, and formulate the best scheduling strategy based on traffic prediction and positioning results; Step 5. Use traffic regularization methods to optimize data rate and packet size to improve data transmission efficiency; Step 6. Deploy traffic marking and priority mechanisms to ensure low-latency flows; Step 7. Realize collaborative scheduling optimization of heterogeneous devices based on reinforcement learning.

[0091] These seven steps constitute the core content of the present invention. First, the purpose of establishing a traffic relationship model is to provide data support for subsequent optimization decisions. For example, in the traffic management system of a smart city, by analyzing historical traffic flow data, we can predict future road congestion and make corresponding traffic signal adjustments in advance. Secondly, achieving seamless connection of network hardware interfaces can adapt to different types of edge devices and improve the compatibility and flexibility of the system. For example, when a new traffic camera is added to the network, the system can automatically identify and configure the appropriate communication protocol without manual intervention.

[0092] Then, the advantage of constructing a topology mapping table is that it can reflect the structural changes of the network in real time and provide accurate information for subsequent traffic scheduling. In the industrial Internet of Things scenario, when a sensor node on a production line suddenly fails, the system can quickly adjust the data routing strategy to ensure the continuity and stability of production monitoring. Next, the traffic scheduling rules based on reinforcement learning are an innovation of the present invention. Through machine learning algorithms, the system can automatically optimize the scheduling strategy based on historical data and current network conditions. For example, in Internet of Vehicles applications, the system may prioritize the data transmission of emergency vehicles during peak traffic hours, and postpone non-urgent entertainment data to the trough period.

[0093] Again, the importance of traffic regularization methods lies in that they can reduce network congestion and improve bandwidth utilization. In an edge computing environment, by dynamically adjusting the size of data packets, the system can reduce network load while ensuring transmission efficiency. In addition, traffic marking and priority mechanisms can ensure the service quality of key businesses. For example, in the scenario of a smart factory, the transmission of control instructions may be given the highest priority to ensure real-time response of the production line. Finally, the collaborative scheduling optimization of heterogeneous devices based on reinforcement learning fully reflects the intelligence and adaptability of the present invention. Through continuous learning and optimization, the system can make the best resource allocation decisions in a complex network environment. For example, when the edge computing node is overloaded, the system may transfer part of the computing tasks to nearby nodes or the cloud to balance the overall performance.

[0094] Next, the detailed implementation of step 1 includes establishing a data traffic relationship model between the storage network and the computing network based on the timestamp and address information of the data packet. Specifically, for the storage network part, we established an equation for the storage capacity of the storage network to change over time. The storage capacity change rate function is denoted as H. According to the storage capacity change rate function calculation formula, the change of storage network storage capacity over time can be expressed as: Where H(t) = β·e -λt ·B·V.

[0095] In this formula, t represents any time, S(t) represents the storage capacity at time t, S0 represents the initial storage capacity, β and λ represent storage change rate parameters, B represents the data block size, and V represents the average data write rate. The advantage of this model is that it can accurately predict the capacity changes of the storage network, thereby better managing data storage and transmission. For example, in the edge nodes of a large data center, we can predict the change trend of storage capacity based on this model, adjust the data distribution strategy in time, and avoid storage overflow in some nodes.

[0096] For the computational network part, we introduce the concepts of throughput T and average link utilization U. The calculation formula for average link utilization is: Where L is the length of each packet, T is the throughput, B is the bandwidth capacity, and N is the number of links on each path. The relationship between the throughput and utilization of each path is: The advantage of this model is that it can help us identify bottlenecks in the network and optimize them accordingly.

[0097] Finally, we get the relationship between data transfer rate and computing time and storage time: Among them, α, β, and γ represent the parameters of the throughput change rate. This comprehensive model takes into account multiple factors such as network utilization, bandwidth, packet size, and storage rate, and can fully reflect the operating status of the network and provide strong support for subsequent optimization decisions.

[0098] Next, the implementation of step 2 is to achieve seamless connection of network hardware interfaces and deploy adaptive bridging methods. First, we define the timestamp and transmission rate of data packets and calculate the expected value of the data flow length in the network: where T1, T2, T3, ..., S n It is represented as the timestamp of the data packet, B is the length of each packet, and X is the length of the data stream.

[0099] Then, based on the adaptive bridging method, we complete the seamless connection of the network hardware interface. This process includes defining the total number of transmission nodes K, the edge B between nodes i and j i，j The bandwidth capacity of the link connecting nodes i and j. We introduce the concept of traffic marking weight value: Where T i Represents the throughput of node i. The introduction of this weight value enables the system to distribute traffic more intelligently.

[0100] When judging the congestion of the current network traffic at each node, we adopt a dynamic routing strategy. If the current network traffic value is greater than the current network state, node i selects the node with the current traffic less than the network state B. i，jThe node of is taken as the next hop node, and the next node of node j is calculated: Next(j) = argmax k∈N(j) (B j，k -F j，k ), where Next(j) represents the next node of the node, N(j) represents the set of neighbor nodes of the node, and B j，k represents the link bandwidth capacity from node to node k, F j，k represents the current flow from node j to node k. This method can effectively avoid network congestion and improve the efficiency of data transmission.

[0101] If the node is the destination node, the data reaches the destination node through the data link, and the data packet is passed to the storage network resources for storage or the computing network resources for calculation; if the node is not the destination node, the data packet will be sent to the next hop node for the next iteration; if the next hop node does not exist, it returns to node i and broadcasts a data retransmission message to supplement the link's load capacity.

[0102] Finally, we select the best bridge point to complete the seamless connection of the network. The advantage of this adaptive bridging method is that it can dynamically adjust the data transmission path according to the real-time status of the network, improving the flexibility and robustness of the network. For example, in an IoT system in a large industrial park, when some network links are congested, the system can automatically select alternative paths to ensure the timely transmission of critical production data.

[0103] Next, the specific implementation of step 3 is to build a topology mapping table and use storage resources to dynamically manage analysis. First, we define the input flow I of the computing network node k k and output flow O k , and then proposed a method to calculate the flow of node k in the network:

[0104]

[0105] In this formula, k represents the node, F k represents the traffic of node k, I k represents the input flow of node k, O k represents the output flow of node k, N(k) represents the set of neighbor nodes of node k, and F m，k represents the traffic from the previous hop node m to node k, F k，n It represents the traffic that enters node k and reaches the next hop n of node k within a period of time. The advantage of this model is that it can comprehensively evaluate the load of each node in the network and provide a basis for subsequent optimization decisions.

[0106] For example, in a smart city video surveillance system, we can use this model to evaluate the data processing load of each edge node. When the system detects that the load of a node is too high, it can adjust the data processing strategy in time and transfer some tasks to nodes with lower load, thus ensuring the smooth operation of the entire system.

[0107] Then, when the traffic between nodes in the computing network meets the transmission rules, we connect the computing network nodes and the storage network nodes and establish a topology mapping table. The purpose of this mapping table is to provide the system with a clear view of the network structure, which helps to allocate resources and schedule tasks more efficiently.

[0108] Next, the detailed implementation of step 4 is to formulate traffic scheduling rules based on reinforcement learning. First, we introduced a threshold function in the storage network to manage the consumption of computing resources. The expression of this threshold function is:

[0109]

[0110] in, represents the output value of the threshold function, α and β are the parameters of the threshold function, and x represents the input data. The introduction of this threshold function enables the system to manage computing resources more intelligently. For example, in an edge computing node, we can set α = 0.8 and β = 0.2. This means that when the resource utilization rate exceeds 80%, the system will take measures to limit the access of new tasks; when the utilization rate is less than 20%, the system will actively accept new tasks; between 20% and 80%, the system will make decisions based on the specific situation.

[0111] Then, within the time threshold, we analyze the dynamic changes of network resources and, based on the predicted values ​​of the traffic relationship model, we derive the bottleneck of the storage network and implement traffic allocation. The advantage of this traffic scheduling method based on reinforcement learning is that it can continuously optimize the scheduling strategy based on historical data and current network conditions, thereby improving the adaptability and efficiency of the system.

[0112] For example, in an automated system of a large logistics warehouse, we can use this method to optimize data transmission and processing within the warehouse. The system will predict future order volumes based on historical order processing data and current inventory status, and adjust data processing priorities and resource allocation accordingly to ensure that all logistics information can be processed in a timely manner during peak order periods.

[0113] Next, in one embodiment of the present invention, the specific implementation of step 5 is a method for optimizing the data rate and packet size using a traffic regularization method. First, we propose a method for calculating the transmission time: Where X and Y represent the total data rate and total packet size of the transmission, M represents the packet size, and N represents the number of links in the network.

[0114] Then, we use the exponential weighted average algorithm to calculate the adjustment ratio of the transmission rate. Specifically, the average exponential weight T is used as the basis for the adjustment ratio, and its calculation formula is:

[0115] T(t)=β·T(t-1)+(1-β)·V(t)

[0116] Where β represents the weight coefficient and V(t) represents the current data rate at time t. The advantage of this algorithm is that it can smooth short-term fluctuations while being sensitive to long-term trends. In practical applications, we can usually set β to 0.9, which can effectively reduce the impact of noise while remaining sensitive to the latest data.

[0117] Finally, we calculate the average value W of the adjusted data transmission rate based on the calculated average exponential weight T and the dynamically adjusted data transmission speed rate V. The formula is as follows: W = β·T+(1-β)·V. The advantage of this traffic regularization method is that it can dynamically adjust the data transmission rate, which not only ensures transmission efficiency but also avoids network congestion.

[0118] For example, in a production line monitoring system in a smart factory, we can use this method to optimize the transmission of sensor data. When the production line is running normally, the system may reduce the frequency of data transmission to reduce the network load; when an abnormality is detected, the system will immediately increase the data transmission rate to ensure that managers can obtain detailed production status information in a timely manner.

[0119] In one embodiment of the present invention, a specific method of deploying a traffic marking and priority mechanism is provided. First, a priority marking field is added to the header of a transmission frame, and different priority markings are assigned to data streams. The purpose of this is to provide differentiated quality of service guarantees for different types of data streams.

[0120] Then, we use the scheduler to calculate the remaining time (Time To Go, TTG) of each data flow in the queue. The calculation formula is as follows: Where TTG represents the remaining time of each data flow in the queue, L d Indicates the target node of the data flow, L c Indicates the current node, O k represents the output flow of node k, L represents the length of the data flow, and M represents the length of each packet. This formula takes into account multiple factors such as the destination of the data flow, the current location, and the network status, and can more accurately estimate the time required for data transmission.

[0121] Finally, when sending data streams to queue, the system will send them in the order of the queue according to the data stream priority tag. The advantage of this mechanism is that it can ensure that high-priority data streams (such as emergency instructions, real-time monitoring data, etc.) are processed first, thereby improving the response speed and reliability of the system.

[0122] For example, in a smart medical system, we can use this mechanism to optimize the transmission of different types of medical data. We can give the highest priority to the real-time location information of emergency ambulances and patient vital signs data to ensure that these critical information can be transmitted to the hospital immediately; while for daily health monitoring data, we can give it a lower priority and transmit it when the network load is lighter.

[0123] In one embodiment of the present invention, the coordinated scheduling optimization of heterogeneous devices based on reinforcement learning is implemented. First, we model the heterogeneous devices, match the heterogeneous devices with the storage network and the computing network, and assign system parameters to the heterogeneous devices. These parameters include memory, CPU speed, bandwidth, power consumption, storage resources, and data flow.

[0124] Then, each network traffic node of the network device is trained through heterogeneous devices to obtain the traffic scheduling results. This process uses machine learning technology, enabling the system to automatically optimize the scheduling strategy based on historical data and current status.

[0125] Next, we set up a reward function to achieve optimal policy learning. Specifically, we introduced a reward function T based on the network delay of the storage device. reward and a reward function C based on the network delay of the computing device reward Their calculation formulas are: T reward =-W s ·T s and C reward =-W c ·T c , where T s and T c represents the storage network and computing network delay, W s and W c is represented as an optimizable parameter.

[0126] By selecting the corresponding reward function through system parameters, we distribute traffic to the storage network and computing network, and use the optimized parameter W as the benchmark to obtain the maximum reward function Q i :

[0127]

[0128] In this formula, R krepresents the reward of network k, γ represents the discount factor (usually set to 0.95 to balance short-term and long-term benefits), s′ represents the next state, and a represents the action.

[0129] Finally, we use the Q table in reinforcement learning to iterate until the maximum number of iterations is reached to obtain a trained network device. After the training is completed, the system can automatically schedule computing tasks and storage tasks to the most suitable edge device based on real-time conditions.

[0130] The advantage of this reinforcement learning-based optimization method for collaborative scheduling of heterogeneous devices is that it can adapt to complex and changing network environments, continuously learn and improve scheduling strategies, and thus achieve optimal resource allocation. For example, in a large-scale smart city management system, we can use this method to optimize data processing and resource allocation between various subsystems of the city (such as transportation, energy, security, etc.). The system will dynamically adjust the allocation of computing tasks according to the real-time needs and performance status of each subsystem to ensure the efficient operation of the entire city management system.

[0131] The present invention also discloses an adaptive network traffic optimization system based on edge computing. The system includes a storage network 1, a computing network 2 and an edge device 3. The edge device 3 includes a traffic regularization module 31, a traffic marking module 32 and a traffic scheduling module 33. The storage network 1 includes a storage network device, and the computing network 2 includes a computing network device.

[0132] In this system, the edge device 3 is connected to the storage network device and the computing network device through wired or wireless connections, and the storage network device and the computing network device are also connected through wired or wireless connections. The traffic between the edge device 3 and the storage network device and the computing network device is processed by the traffic shaping module 31, the traffic marking module 32 and the traffic scheduling module 33 before transmission.

[0133] The advantage of this system architecture is that it makes full use of the characteristics of edge computing and delegates data processing and decision-making processes to places close to the data source, thereby greatly reducing network latency and improving the response speed and efficiency of the system. At the same time, through the coordinated work of the traffic shaping module 31, the traffic marking module 32 and the traffic scheduling module 33, the system can manage network resources more intelligently and adapt to various complex network environments.

[0134] For example, in an intelligent transportation system, we can deploy edge devices 3 at each major intersection, which can process data from traffic cameras, vehicle-mounted sensors, etc. in real time. Through the traffic shaping module 31, the system can adjust the data transmission rate according to the current network conditions; through the traffic marking module 32, the system can give higher priority to the data of emergency vehicles; through the traffic scheduling module 33, the system can dynamically adjust the timing of traffic lights according to the real-time conditions of each intersection. In this way, the entire transportation system can operate more intelligently and efficiently.

[0135] First, storage network devices are defined as intelligent devices based on cloud computing and distributed storage methods. These devices are accessed in the form of data streams, and flexible scheduling of computing tasks and storage tasks is achieved through storage resource management. The advantage of this design is that it can make full use of the powerful computing power of cloud computing and the high reliability of distributed storage to provide stable backend support for the entire system.

[0136] Secondly, computing network devices are defined as intelligent devices based on machine learning methods. These devices are also accessed in the form of data streams, and through computing resource management, flexible scheduling of computing tasks and storage tasks is achieved. The advantage of this design is that it can use machine learning technology to continuously optimize the system's operating strategy and improve the system's intelligence level and adaptability.

[0137] Finally, edge devices 3 are defined as intelligent devices based on edge computing technology. These devices are accessed in the form of data streams, and load balancing of low-latency data is achieved by running reinforcement learning and traffic management. The advantage of this design is that it can quickly process data at the edge of the network, reduce network transmission delays, and improve the real-time performance of the system.

[0138] For example, in a smart manufacturing system in a large industrial park, we can use this system architecture to optimize the production process. Storage network devices can be used to store and manage massive production data and product design files; computing network devices can be used to run complex production optimization algorithms and quality control models; edge devices 3 can be deployed on various production lines to process data from various sensors in real time and make control decisions quickly. Through the collaborative work of these three devices, the entire manufacturing system can achieve a more efficient, flexible, and intelligent production process.

[0139] In general, the adaptive network traffic optimization method based on edge computing and its system proposed in the present invention have the following significant advantages: first, it can adapt to changes in the network environment in real time and improve the flexibility and robustness of the system; second, through intelligent traffic scheduling and resource allocation, it can significantly improve network performance and reduce latency; third, the optimization strategy based on reinforcement learning enables the system to continuously improve itself and adapt to various complex network scenarios; finally, it makes full use of the advantages of edge computing and provides strong support for future Internet of Things and 5G applications. This method and system have broad application prospects in many fields such as smart cities, industrial Internet of Things, and intelligent transportation, and are expected to provide innovative solutions to the challenges brought about by the current surge in network traffic.

[0140] In order to verify the superiority of the adaptive network traffic optimization method based on edge computing and its system, we designed a set of comparative experiments. In this experiment, we selected a typical smart city scenario as the test environment, specifically involving three main aspects: traffic management, environmental monitoring, and public safety.

[0141] Example 1 adopts the method and system of the present invention, including a storage network 1, a computing network 2 and an edge device 3, wherein the edge device 3 is equipped with a traffic shaping module 31, a traffic marking module 32 and a traffic scheduling module 33. In contrast, Comparative Example 1 adopts a traditional centralized cloud computing architecture, and all data processing and decision-making are performed in a remote data center.

[0142] We selected four key indicators to evaluate system performance: average response time, network bandwidth utilization, system reliability, and energy efficiency. The selection of these indicators fully reflects the innovation and advantages of this invention. The specific detection standards and methods are as follows:

[0143] 1. Average response time: We simulated 1,000 emergency events (such as traffic accidents, sudden environmental pollution, etc.) and measured the average time from the occurrence of the event to the system's response. The unit is milliseconds (ms).

[0144] 2. Network bandwidth utilization: Under normal operating conditions, measure the average utilization of network bandwidth within 24 hours. Expressed as a percentage.

[0145] 3. System reliability: We conducted a 30-day continuous operation test and recorded the number and duration of system failures. The system uptime is expressed as a percentage of the total time.

[0146] 4. Energy efficiency: Measures the average energy consumption of the entire system over a 7-day period. The unit is kilowatt-hour (kWh).

[0147] After a month of rigorous testing, we obtained the following results:

[0148] Test indicators Example 1 Comparative Example 1 Average response time (ms) 50 200 Network bandwidth utilization (%) 75 45 System reliability (%) 99.99 99.9 Energy efficiency (kWh) 500 800

[0149] From these test results, we can clearly see the superiority of the present invention. First, in terms of average response time, the method of the present invention only takes 50 milliseconds, while the traditional method takes 200 milliseconds. This significant improvement stems from the edge computing architecture adopted by the present invention, which delegates data processing and decision-making processes to places close to the data source, greatly reducing network transmission delays. In the scenario of a smart city, this means that emergencies such as traffic accidents and environmental pollution can be responded to and handled faster, potentially saving lives and reducing losses.

[0150] Secondly, the present invention also performs well in terms of network bandwidth utilization, reaching 75%, which is much higher than the 45% of the traditional method. This is due to the collaborative work of the traffic shaping module 31 and the traffic scheduling module 33 of the present invention, which can dynamically adjust the data transmission strategy according to the real-time network status. Efficient bandwidth utilization not only improves the overall efficiency of the system, but also leaves ample room for future expansion.

[0151] In terms of system reliability, the present invention has achieved a high level of 99.99%, which is a further improvement over the 99.9% of traditional methods. This 0.09% difference may seem small, but in the long-term operation of large systems, it means an additional reduction of nearly 8 hours of system downtime per year. This high reliability stems from the distributed architecture and intelligent resource management strategy of the present invention, which effectively reduces the risk of single point failures.

[0152] Finally, in terms of energy efficiency, the advantages of the present invention are also very obvious. During the 7-day test period, the system of the present invention consumed only 500 kWh of electricity, while the traditional method consumed 800 kWh. This 37.5% energy saving effect not only brings considerable economic benefits, but also meets the requirements of today's society for green environmental protection. This high energy efficiency is mainly due to the edge computing architecture of the present invention, which reduces the remote transmission and processing of large amounts of data, as well as the intelligent resource scheduling strategy based on reinforcement learning.

[0153] In summary, Example 1 is undoubtedly the best embodiment of the present invention. It is significantly superior to traditional methods in many aspects such as response speed, resource utilization, reliability and energy efficiency. These advantages fully demonstrate the great potential of the adaptive network traffic optimization method based on edge computing proposed in the present invention and its system in practical applications. Especially in complex smart city scenarios, the present invention can provide faster response, more efficient resource utilization, more reliable system operation and lower energy consumption. This can not only improve the efficiency of urban management, but also provide citizens with a better service experience while reducing operating costs and environmental impact.

[0154] It is worth mentioning that although this test is mainly aimed at smart city scenarios, the advantages of the present invention are not limited to this. In fact, this adaptive network traffic optimization method based on edge computing and its system have great application prospects in many fields such as industrial Internet of Things, intelligent transportation, and telemedicine. With the widespread deployment of 5G technology and the increasing number of Internet of Things devices, the method and system proposed in the present invention will play an increasingly important role in the future intelligent and networked world.

[0155] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. The adaptive network traffic optimization method based on edge computing is characterized by The following steps are involved: Step 1. Establish a traffic relationship model to analyze the status of data flow in the storage network and computing network; Step 2. Realize seamless connection of network hardware interfaces, deploy adaptive bridging methods, and complete network communication; Step 3. Construct a topology mapping table, and use dynamic management and analysis of storage resources to provide a basis for optimized scheduling; Step 4. Formulate traffic scheduling rules based on reinforcement learning, and formulate the best scheduling strategy based on traffic prediction and positioning results; Step 5. Use traffic regularization methods to optimize data rate and packet size to improve data transmission efficiency; Step 6. Deploy traffic marking and priority mechanisms to ensure low-latency flows; Step 7. Implement collaborative scheduling optimization of heterogeneous devices based on reinforcement learning.

2. The method for adaptive network traffic optimization based on edge computing according to claim 1 is characterized in that: The detailed steps of step 1 are as follows: 1-1. Based on the timestamp and address information of the data packet, a data traffic relationship model in the storage network and computing network is established, including: a. For the storage network part, establish the equation of storage network storage capacity changing over time. The storage capacity change rate function is recorded as H. According to the storage capacity change rate function calculation formula, the storage network storage capacity changing over time can be expressed as: H(t)=β·e -λt ·B·V Where t represents any time, S(t) represents the storage capacity at time t, S0 represents the initial storage capacity, H(τ) represents the storage capacity change rate function, β and λ represent storage change rate parameters, B represents the data block size, and V represents the average data write rate; b. For the calculation network part, establish an expression for calculating the change of network throughput over time. The average link utilization is: Where L represents the length of each packet, T represents the throughput, B represents the bandwidth capacity, and N represents the number of links on each path; The relationship between the throughput and utilization of each path is: According to the above steps, the relationship function between data transmission rate and computing time and storage time can be expressed as: Wherein, R(t) represents the data transmission rate at time t, α, β, γ represent the parameters of the throughput change rate, and V represents the average data writing rate.

3. The method for adaptive network traffic optimization based on edge computing according to claim 1 is characterized in that: The detailed steps of step 2 are as follows: 2-1. Define the timestamp of the data packet. Assume that the timestamp of the data packet is T1, T2, T3, ..., S n , the transmission rate is recorded as A, the size of the data flow in the network is recorded as B, and the expected value of the data flow length in the network is: Among them, T1, T2, T3, ...S n It is represented as the timestamp of the data packet, B is the length of each packet, X is the length of the data stream, and E(X) is the expected value of the length of the data stream; 2-2. Based on the adaptive bridging method, the seamless connection of the network hardware interface is completed, and the steps include: a. Define the total number of transmission nodes as K, and the edge between nodes i and j as B i，j is the link bandwidth capacity connecting nodes i and j, denoted as B i，j =w ij , the throughput of node i is denoted as T i , its traffic marking weight value is: Among them, W i represents the traffic marking weight value of node i, T i represents the throughput of node i, K represents the total number of transmission nodes; b. Determine the congestion of the current network traffic at each node: compare the traffic size with the current network status. If the current network traffic value is less than or equal to the current network status, no data packets will be sent; if the current network traffic value is greater than the current network status, node i will select the current traffic value less than the network status B. i，j The node of is taken as the next hop node, and the next node of node j is calculated: Next(j)=argmax k∈N(j) (B j，k -F j，k ) Among them, Next(j) represents the next node of the node, N(j) represents the set of neighbor nodes of the node, and B j，k represents the link bandwidth capacity from node to node k, F j，k Represents the current flow from node j to node k; If the node is the destination node, the data reaches the destination node through the data link, and the data packet is passed to the storage network resource for storage or the computing network resource for calculation; if the node is not the destination node, the data packet will be sent to the next hop node for the next iteration; if the next hop node does not exist, it returns to node i and broadcasts a data retransmission message to supplement the link load capacity; 2-3. Select the best bridge point to complete seamless network connection.

4. The method for adaptive network traffic optimization based on edge computing according to claim 1, characterized in that: The detailed steps of step 3 are as follows: 3-1. Define the input flow of computing network node k as I k , the output flow is O k , then the flow of node k in the computing network can be calculated as: Among them, k represents the node, F k represents the traffic of node k, I k represents the input flow of node k, O k represents the output flow of node k, N(k) represents the set of neighbor nodes of node k, and F m，k represents the traffic from the previous hop node m to node k, F k，n It represents the traffic that enters node k and reaches the next hop n of node k within a period of time; 3-2. When the traffic between nodes in the computing network meets the transmission rules, the computing network nodes and the storage network nodes are connected, and a topology mapping table is established.

5. The method for adaptive network traffic optimization based on edge computing according to claim 1, characterized in that: The detailed steps of step 4 are as follows: 4-1. In the storage network, since computing resources need to be consumed, a threshold function is set, which is expressed as: Among them, α, β represent the parameters of the threshold function, x represents the input data, Represents the output value of the threshold function; 4-2. Within the time threshold, analyze the dynamic changes of network resources, and based on the predicted value of the traffic relationship model, obtain the bottleneck of the storage network and realize the traffic distribution.

6. The method for adaptive network traffic optimization based on edge computing according to claim 1, characterized in that: The detailed steps of step 5 are as follows: 5-1. Weighting the data rate and packet size, the transmission time is expressed as: Where X and Y represent the total data rate and total packet size of the transmission, T represents the transmission time, M represents the packet size, and N represents the number of links in the network; The transmission rate is calculated using an exponential weighted average algorithm, and the average exponential weight T is used as the basis for adjusting the ratio. The formula can be expressed as: T(t)=β·T(t-1)+(1-β)·V(t) Where β represents the weight coefficient, T(t) represents the reference of the adjustment ratio at time t, and V(t) represents the current data rate at time t; 5-2. According to the average index weight T calculated in 5-1 and the dynamically adjusted data transmission speed rate V, calculate the average value W of the adjusted data transmission rate. The formula is as follows: W=β·T+(1-β)·V Wherein, W represents the average value of the adjusted data transmission rate, V represents the current data rate, T represents the adjustment ratio, and β represents the weight coefficient.

7. The method for adaptive network traffic optimization based on edge computing according to claim 1, characterized in that: The detailed steps of step 6 are as follows: 6-1. Add a priority tag field to the header of the transmitted frame and assign different priority tags to the data streams; 6-2. The scheduler calculates the remaining time (Time To Go, TTG) of each data flow in the queue. The formula is as follows: Among them, TTG represents the remaining time of each data flow in the queue, L d represents the target node of the data flow, L c Indicates the current node, O k represents the output flow of node k, L represents the length of data flow, and M represents the length of each packet; 6-3. When sending data streams to queue, they are sent in the order of the queue according to the data stream priority tag.

8. The method for adaptive network traffic optimization based on edge computing according to claim 1, characterized in that: The detailed steps of step 7 are as follows: 7-1. Model heterogeneous devices and match them with storage networks and computing networks; at the same time, assign system parameters to heterogeneous devices; The system parameters include: memory, CPU speed, bandwidth, power consumption, storage resources and data flow; 7-2. Each network traffic node of the network device is trained through heterogeneous devices to obtain traffic scheduling results; 7-3. Set the reward function to achieve optimal strategy learning: the reward function includes a network delay reward function T based on the storage device reward and a reward function C based on the network delay of the computing device reward for: T reward =-W s ·T s C reward =-W c ·T c Among them, T s and T c represents the storage network and computing network delay, W s and W c Expressed as an optimizable parameter; select the corresponding reward function through the system parameters, distribute the traffic to the storage network and the computing network, and use the optimized parameter W as the benchmark to obtain the maximum reward function Q i : Among them, R k represents the benefit of network k, γ represents the discount factor, s′ represents the next state, a represents the action, K represents the network set, A represents the action set, and Q(s′, a) represents the state-action value function; 7-4. Use the Q table in reinforcement learning to iterate until the maximum number of iterations is reached to obtain a trained network device; 7-5. Trained network devices schedule computing tasks and storage tasks to edge devices.

9. A system for executing the edge computing-based adaptive network traffic optimization method according to any one of claims 1 to 8, characterized in that: It includes a storage network, a computing network and an edge device, wherein the edge device includes a traffic shaping module, a traffic marking module and a traffic scheduling module. The storage network includes a storage network device, and the computing network includes a computing network device. The edge device is connected to the storage network device and the computing network device through a wired connection or a wireless connection, respectively, and the storage network device and the computing network device are connected through a wired connection or a wireless connection. The traffic between the edge device and the storage network device and the computing network device is transmitted after being processed by the traffic shaping module, the traffic marking module and the traffic scheduling module.

10. The adaptive network traffic optimization system based on edge computing according to claim 9, characterized in that: The storage network device is an intelligent device based on cloud computing and distributed storage methods, which is accessed in the form of data streams, and flexible scheduling of computing tasks and storage tasks is achieved through storage resource management; the computing network device is an intelligent device based on machine learning methods, which is accessed in the form of data streams, and flexible scheduling of computing tasks and storage tasks is achieved through computing resource management; the edge device is an intelligent device based on edge computing methods, which is accessed in the form of data streams, and load balancing of low-latency data is achieved through running reinforcement learning and traffic management.

Images