Comprehensive congestion control and adaptive transmission optimization method based on WAPI

Abstract

The invention provides a comprehensive congestion control and adaptive transmission optimization method based on WAPI, and belongs to the field of communication. The key problems that under the wireless communication environment based on the WAPI security standard in the electric power Internet of Things, network congestion is difficult to dynamically regulate and control, data transmission is prone to disorder, and a retransmission mechanism lacks intelligent adaptation are solved, and the method comprises the following steps that WAPI environment initialization and security association establishment are carried out; based on a WAPI and reinforcement learning congestion control framework, a teacher-student distillation mechanism is adopted to extract a symbol strategy, and multi-context branch agents are integrated to adapt to different network conditions, so that long-term congestion planning and dynamic rate adjustment are realized; optimizing a data packet sequence and congestion window growth coupling in multi-path transmission through ECN marking and correlation verification; running a WAPI security aware adaptive retransmission module, and dynamically selecting unicast or multicast retransmission in combination with a WAPI security mechanism; the present application is applied to WAPI.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication control, and in particular relates to a comprehensive congestion control and adaptive transmission optimization method based on WAPI. BACKGROUND

[0002] With the rapid development of power Internet of Things (Power IoT), a large number of terminal devices access wireless networks, and the types of services are continuously expanding from real-time monitoring, control and scheduling to video transmission. The complexity and dynamicity of network structure are significantly improved, and higher requirements are put forward for the security, reliability and adaptability of the communication system. As a wireless security standard independently developed by China, WAPI (Wireless Local Area Network Authentication and Privacy Infrastructure) has gradually become an important security guarantee for power communication systems due to its reliable authentication and strong security. However, in the WAPI network environment, the wireless channel has characteristics such as large bandwidth fluctuation, strong randomness of packet loss, and unstable delay, which makes it difficult for traditional network optimization and congestion control methods to achieve efficient transmission while ensuring security. Especially in the power Internet of Things scenario with multiple user access and multi-path transmission, network bottleneck competition and out-of-order phenomenon frequently occur, which directly affects the real-time response of critical services and system reliability.

[0003] In existing research, traditional congestion control algorithms such as TCP CUBIC and TCP BBR mainly adjust the rate based on packet loss or delay signals, and perform well in stable networks, but lack flexibility in dynamic and heterogeneous wireless environments such as power Internet of Things, and are difficult to cope with random packet loss and burst congestion, resulting in a decrease in bandwidth utilization. In recent years, deep reinforcement learning such as the Aurora algorithm uses neural networks to achieve adaptive optimization and can improve throughput. However, this type of method has the problems of uninterpretable decision-making process, high reasoning delay and large computational overhead, and is not suitable for deployment in resource-constrained power terminals. To enhance interpretability, symbolic regression and policy distillation methods are proposed to convert deep model strategies into symbolic decision trees to achieve lightweight reasoning, but existing schemes are mostly static rules and are difficult to adapt to complex network scenarios with sudden changes in bandwidth or changes in packet loss types.

[0004] In terms of multi-path scheduling, traditional algorithms such as LowRTT and FPS mainly rely on delay prediction for scheduling, but ignore the shared bottleneck characteristics among multiple sub-flows. When multiple sub-flows compete for the same link and lack of coordinated control, it is easy to cause serious congestion and out-of-order at the receiving end, increasing the cache pressure and reducing the real-time performance. In terms of transmission reliability, some existing studies use static unicast or multicast retransmission mechanisms, or use deep learning to predict packet loss events, but such methods either consume a lot of resources and have low energy efficiency, or are not combined with WAPI security mechanisms, making it difficult to meet the comprehensive needs of security, real-time performance and energy efficiency of power Internet of Things.

[0005] In summary, existing congestion control algorithms often cannot accurately distinguish between random packet loss and congestion packet loss when facing complex and variable wireless links, resulting in transmission rate adjustment lag or over-regulation, thereby reducing bandwidth utilization. Although deep reinforcement learning-based algorithms can improve performance, they have the problems of uninterpretable model decision-making process, high computational complexity, and difficulty in real-time deployment on resource-constrained terminal devices. At the same time, in the multi-path transmission scenario, traditional scheduling strategies do not consider the shared bottleneck characteristics of multiple sub-flows in the WAPI network, which can easily cause path competition to intensify and out-of-order accumulation at the receiving end, seriously affecting the real-time performance and stability of power services. In addition, existing retransmission mechanisms mostly use fixed unicast or multicast methods, which cannot adaptively switch according to network load and terminal state, and are not deeply integrated with WAPI security authentication mechanisms, resulting in resource waste and data security risks. That is, existing technologies cannot simultaneously achieve secure and reliable, lightweight and efficient, and dynamic adaptive communication control in the WAPI environment. SUMMARY

[0006] To solve the key problems of network congestion being difficult to dynamically control, data transmission being prone to out-of-order, and retransmission mechanisms lacking intelligent adaptation in the wireless communication environment based on the WAPI security standard in the power Internet of Things, the present application proposes a comprehensive congestion control and adaptive transmission optimization method based on WAPI, which builds a comprehensive communication optimization method integrating congestion control, out-of-order suppression, and adaptive retransmission under the premise of ensuring the WAPI security authentication system, and realizes dynamic transmission control with high security, high reliability, and high efficiency in the power Internet of Things.

[0007] The technical solution adopted by the present application is: a comprehensive congestion control and adaptive transmission optimization method based on WAPI, comprising the following steps:

[0008] Step 1, WAPI environment initialization and security association establishment are performed;

[0009] Step 2, on the basis of security initialization, deploy a congestion control module based on symbolic distillation, which is deployed in a simulation platform simulating WAPI network characteristics, collects decision trajectories in multiple typical scenarios, forms a dataset containing WAPI statistics, and converts the initial strategy into an interpretable symbolic strategy through symbolic distillation, introduces WAPI-specific operators, and uses genetic programming optimization rules to generate lightweight control strategies that adapt to different WAPI network contexts;

[0010] Step 3, start the data packet scheduling module based on ECN and forward prediction, which configures ECN queue threshold according to WAPI QoS level, receives feedback of congestion marker ECE through WAPI encrypted channel, detects shared bottleneck subflow based on ECE signal, predicts the transmission volume of each subflow within half RTT based on WAPI link RTT data, dynamically adjusts the sending position of slow subflow, and real-time corrects the prediction parameters through WAPI link monitoring;

[0011] Step 4, run the WAPI security-aware adaptive retransmission module, AC obtains encrypted network parameters in real time through WAPI security channel, calculates the current network load and link utilization, and makes retransmission decision based on dynamic threshold: when the load is low, use WAPI unicast security channel to retransmit to the packet loss terminal; when the load is high, select unicast or multicast mode with smaller expected load according to the number of packet loss terminals, and all retransmission data are encrypted through WAPI.

[0012] Further, the congestion control module based on symbolic distillation is integrated with a congestion control framework based on WAPI and reinforcement learning, which extracts symbolic strategies using a teacher-student distillation mechanism, and integrates multiple context branch agents to adapt to different network conditions, realizes long-term congestion planning and dynamic rate adjustment, and the initial strategy in step 2 is the original congestion control strategy in the form of deep neural network.

[0013] Further, the congestion control framework based on WAPI and reinforcement learning also sets a WAPI branch reinforcement learning framework, which first divides the WAPI network conditions into 3 non-overlapping contexts, then trains a special branch agent for each context, and the training data comes from the teacher trajectory subset of the corresponding scene; finally, through K-means clustering analysis of the reward data of each scene, the branch boundary conditions are automatically determined.

[0014] Further, PPO algorithm is used to train the teacher agent in WAPI scenario, cross-entropy loss is used as feedback signal to realize mutation, crossover and pruning of rules, and the optimal symbolic strategy is selected, and the cross-entropy loss of symbolic strategy is The formula is defined as:

[0015] ;

[0016] In the formula: Teacher decision tag, Indicates that the rate needs to be increased, Indicates that the rate needs to be reduced / maintained, Student symbol strategy pair Decision output probability calculated by symbol rule combination, WAPI vector statistical symbol, Total number of trajectory samples in the data set.

[0017] Further, the computer program integrating the packet scheduling method combining network congestion awareness and forward prediction in the ECN and forward prediction-based packet scheduling module has the following implementation steps:

[0018] Step 11: First, determine which data subflows are sharing the same physical bottleneck link, and generate a shared bottleneck candidate set;

[0019] Step 12: Perform congestion event time series correlation analysis on the subflows in the candidate set to confirm that they indeed share the same physical bottleneck link, and form a final shared bottleneck subflow set;

[0020] Step 13: For each round-trip time RTT slower subflow Short fast subflow , when the fast subflow belongs to one of the shared bottleneck candidate sets, the RTT time congestion window is added by 1, and at this time the number of data packets that each fast subflow can send within half the RTT window is calculated as follows:

[0021] ;

[0022] In the formula: represents the congestion window of the subflow ;

[0023] When the fast subflow does not belong to any shared bottleneck candidate set, its behavior is similar to traditional TCP, and the congestion window is added by 1 every RTT window, and at this time the number of data packets that each fast subflow can send within half the RTT window is calculated as follows:

[0024] ;

[0025] Step 14: Add the predicted transmission packet numbers of all subflows faster than the slow subflow to get the total;

[0026] ​Step 15: Treat all data packets to be sent as a logically ordered shared send buffer.

[0027] Furthermore, the WAPI security-aware adaptive retransmission module integrates a power IoT adaptive retransmission control method based on the WAPI protocol, which combines the WAPI security mechanism with dynamic retransmission strategy. By sensing the network status in real time, it adaptively selects unicast or multicast retransmission mode.

[0028] Furthermore, the formulas for calculating the current network load and link utilization in step 4 are as follows:

[0029] ;

[0030] In the formula: For a moment Average network load within, For a moment Average link utilization within the area , They represent the times up to time. The number of multicast and unicast retransmissions that have occurred so far. For link bandwidth, For multicast stream rate, N is the data packet length, and N1 is the power IoT terminal within the base station's coverage area.

[0031] Furthermore, during the retransmission decision-making process, the first step is to check whether the current network load exceeds a set threshold. If the following formula is met, unicast retransmission is prioritized; otherwise, the retransmission is based on the current number of packet-loss terminals. Calculate the expected network load for unicast and multicast modes respectively:

[0032] ;

[0033] In the formula: To dynamically adjust parameters;

[0034] The formulas for calculating the expected network load under unicast and multicast modes are as follows:

[0035] ;

[0036] In the formula: Indicates that in the case that there is known When packet loss occurs at one terminal, and unicast retransmission mode is used, what is the expected normalized load of the network? Indicates that in the case that there is known When multiple terminals experience packet loss, and multicast retransmission mode is used, what is the expected normalized load of the network? This represents the expected number of retransmissions required for each data packet to be successfully transmitted in unicast mode. represents the number of expected retransmission rounds needed to ensure all lost packets are successfully received by all terminals in multicast mode;

[0037] By comparing and , the retransmission mode with smaller expected load is selected.

[0038] Further, in the K-means clustering process, the context center distance criteria formula is defined as:

[0039] ;

[0040] In the formula: is the , class context, is the th feature of , is the th feature of , is the feature weight.

[0041] A computer device comprising a memory, a processor and a computer program stored on the memory, the processor executing the computer program to implement the steps of the method.

[0042] The application has the beneficial effects of the prior art: the comprehensive congestion control and adaptive transmission optimization method based on WAPI provided by the application realizes interpretable lightweight rate control under the premise of guaranteeing WAPI security authentication, combines reinforcement learning and symbol distillation, solves the problems of multi-path competition and out-of-order through the ECN explicit congestion notification and forward prediction mechanism, and designs an adaptive retransmission mechanism that integrates WAPI security perception, which can dynamically switch between unicast and multicast according to the network state, thereby realizing safe, efficient and reliable transmission optimization in the power Internet of Things environment. BRIEF DESCRIPTION OF DRAWINGS

[0043] The application will be further described below in conjunction with the drawings:

[0044] Fig. 1 The structure diagram of the congestion control framework based on WAPI and reinforcement learning provided by the embodiments of the application;

[0045] Fig. 2 The message scheduling method flowchart integrating network congestion perception and forward prediction provided by the embodiments of the application;

[0046] Fig. 3The flowchart illustrates the operation of the adaptive retransmission control method for the power Internet of Things based on the WAPI protocol, provided in this application embodiment, for multi-terminal communication scenarios in smart grids. Detailed Implementation

[0047] like Figs. 1 to 3 As shown, this application provides a comprehensive congestion control and adaptive transmission optimization method based on WAPI. Under the WAPI security authentication framework, this method achieves secure, intelligent and efficient network transmission control through interpretable rate control, multi-path scheduling based on ECN (Explicit Congestion Notification, an improved congestion control method), and adaptive retransmission mechanism.

[0048] The improvements in this application mainly include the following parts:

[0049] 1) A congestion control module based on symbol distillation (SymDistill-CC) was designed. This module integrates a congestion control framework based on WAPI and reinforcement learning, uses a teacher-student distillation mechanism to extract symbol strategies, and integrates a multi-context branch agent to adapt to different network conditions, so as to realize long-term congestion planning and dynamic rate adjustment.

[0050] 2) A packet scheduling module based on ECN and forward prediction (ECN-FPS) was designed. This module integrates a shared bottleneck detection and packet scheduling method based on ECN. By using ECN marking and correlation verification, the coupling between packet order and congestion window growth in multi-path transmission is optimized, avoiding the exacerbation of link congestion by multi-sub-stream competition.

[0051] 3) A WAPI security-aware adaptive retransmission module was designed. This module integrates an adaptive retransmission control method for the power Internet of Things based on the WAPI protocol. It dynamically selects unicast or multicast retransmission in conjunction with the WAPI security mechanism and realizes threshold self-adjustment to balance network load, energy efficiency and service reliability.

[0052] The congestion control framework based on WAPI and reinforcement learning models WAPI congestion control as a sequential decision problem. In each W-MI (WAPI MAC Interval) window, which describes the periodic time window for MAC layer transmission and security scheduling, the agent determines the transmission rate and frame scheduling strategy based on WAPI network statistical observations and calculates the reward. The structure is as follows: Fig. 1 As shown.

[0053] The PPO algorithm is used to train a teacher agent in a WAPI scenario. To address the black-box nature of the teacher model and the decision-making delay in a wireless environment, a teacher-student distillation framework is employed to extract symbolic strategies.

[0054] Firstly, trajectory collection is carried out to collect the deterministic decision trajectory of the teacher agent in 10 typical WAPI scenarios, forming a dataset:

[0055] (1);

[0056] In the formula: WAPI vector statistical symbol, sending rate decision; signal-to-noise ratio at the t time, frame loss rate at the t time, queue length at the t time, service quality level at the t time, wireless resource utilization efficiency at the t time;

[0057] Secondly, the rule space is initialized, the symbol rule search space is defined, and WAPI special operations are added. WAPI special operations are designed for the symbol operators of WAPI protocol characteristics, such as security association state judgment, encrypted frame success rate evaluation, and key update delay calculation.

[0058] Finally, through genetic programming, iteration optimization is carried out, taking cross-entropy loss as the feedback signal to realize mutation, crossover and pruning of rules, and screening the optimal symbol strategy.

[0059] Cross-entropy loss of symbol strategy The formula is defined as:

[0060] (2);

[0061] In the formula: teacher decision label, indicates that the rate needs to be increased, indicates that the rate needs to be reduced / maintained, student symbol strategy decision output probability of calculated by symbol rule combination, for example , Sigmoid function, signal-to-noise ratio threshold function, total number of trajectory samples in the dataset.

[0062] To realize long-term congestion control planning in WAPI scenarios, the agent maintains a set of state attributes across W-MI:

[0063] (3);

[0064] In the formula: the change slope of the frame loss rate in W-MI, for Average signal-to-noise ratio within a W-MI (Wireless Multi-Interface) time window, for The trend of frame loss rate within a W-MI time window for The peak value of the AP cache queue within each W-MI is achieved through a mechanism that avoids the problem of "exponential growth of symbol space due to extended observation history" and enables multi-step planning through attribute transfer.

[0065] To improve the congestion control performance of WAPI in diverse wireless environments, this application proposes a WAPI branch reinforcement learning framework. First, the WAPI network conditions are divided into three non-overlapping contexts. Then, a dedicated branch agent is trained for each context, with training data derived from a subset of teacher trajectories for the corresponding scenario. Finally, K-means clustering analysis is used to automatically determine the branch boundary conditions, avoiding bias from manual partitioning. During clustering, the context center distance criterion is defined as:

[0066] (4);

[0067] In the formula: For the first , Class context, for The One characteristic, for The One characteristic, These are the feature weights.

[0068] Subsequently, to address the packet out-of-order problem caused by shared bottleneck links and path heterogeneity during multi-path transmission in the WAPI power IoT, a packet scheduling method integrating network congestion awareness and forward prediction was designed. Its implementation steps are as follows:

[0069] Step 11: First, determine which data sub-streams are sharing the same physical bottleneck link. For example... Fig. 2 As shown, within a set "observation window", if multiple sub-streams receive ECN tags simultaneously or in a highly synchronized manner, they are initially determined to belong to the same "shared bottleneck candidate set".

[0070] Step 12: Perform more rigorous time series correlation analysis on the sub-flows in the candidate set to confirm that they do indeed share the same bottleneck link, forming the final set of shared bottleneck sub-flows.

[0071] Step 13: For each round-trip time (RTT) compared to the current slow sub-stream Short "tachyon stream" , respectively, calculate the number of data packets it can send in half the RTT window The key difference is whether the fast sub-flow is in the shared bottleneck sub-flow set. If the sub-flow is in the set, its congestion window growth is limited by other shared sub-flows, and grows slowly, in the time, the sub-flow transmits rounds of data, so it needs RTT time to congestion window plus 1, so the number of data packets that can be sent as shown in equation (5).

[0072] (5);

[0073] In the equation: represents the congestion window of the sub-flow .

[0074] When the sub-flow does not belong to any shared bottleneck sub-flow, its behavior is similar to traditional TCP, and the congestion window is added by 1 every RTT window, that is , at this time the number of data packets that can be sent as shown in equation (6).

[0075] (6).

[0076] Step 14: Add the predicted number of packets transmitted by all sub-flows faster than the slow sub-flow (from 1 to ) to get the total .

[0077] Step 15: All data packets to be sent are regarded as a logically ordered shared sending buffer. The sending position of the slow sub-flow is shown in equation (7).

[0078] (7);

[0079] In the equation: is the starting position of the buffer.

[0080] The message scheduling method that integrates network congestion awareness and forward prediction ensures that the expected arrival time of the data packets currently sent by the slow sub-flow will be later than all the data packets sent by the fast sub-flows in the future half RTT. Thus, it ensures that the order of receiving data packets at the receiving end is consistent with the logical order of the sending buffer at the sending end.

[0081] The adaptive retransmission control method based on WAPI protocol fuses the WAPI security mechanism and dynamic retransmission strategy, and significantly optimizes the bandwidth utilization and terminal energy efficiency by real-time sensing of network state, adaptive selection of unicast or multicast retransmission mode. The core includes network state sensing, retransmission decision and threshold dynamic adjustment, and the operation process is as shown in Fig. 3 .

[0082] The retransmission control method takes "security-efficiency-resource" as the unified optimization goal, and solves the rigid problem of traditional retransmission mechanism in power Internet of things. The steps are as follows:

[0083] Step21: In the WAPI architecture, all communication entities need to establish a secure association through certificate authentication and key negotiation to ensure the confidentiality and integrity of data transmission. Suppose there are N1 power Internet of things terminals in the coverage range of a base station, the data packet length is , the link bandwidth is , and the multicast flow rate is set to . These parameters are used as basic input for subsequent state evaluation and decision. The specific principle is as shown in equation (8).

[0084] (8);

[0085] In the formula, is the average network load in time , is the average link utilization rate in time , , respectively represent the number of multicast and unicast retransmissions occurred by time .

[0086] Step22: In the retransmission decision process, first check whether the current network load exceeds the set threshold. If equation (9) is satisfied, prefer to use unicast retransmission to avoid non-packet loss terminals receiving redundant data, reducing link utilization and energy consumption. Otherwise, enter the multi-mode comparison stage.

[0087] (9);

[0088] In the formula, is the dynamic adjustment parameter.

[0089] Step23: If equation (9) is not satisfied, the expected network load under unicast and multicast mode is calculated respectively according to the current number of packet loss terminals . The calculation method is as shown in equation (10).

[0090] (10);

[0091] In the formula: denotes the expected normalized load of the network when unicast retransmission mode is adopted under the condition that there are denotes the expected normalized load of the network when multicast retransmission mode is adopted under the condition that there are denotes the expected number of retransmissions required for each data packet to be successfully transmitted in unicast mode, denotes the expected number of retransmission rounds required to ensure that all

[0092] Step 24: Select the retransmission mode with smaller expected load by comparing and

[0093] The method introduces a dynamic adjustment mechanism for the parameter to adapt to the dynamic environment of the power Internet of Things. According to the real-time network state and service demand, the degree of aggressiveness of the retransmission strategy is flexibly adjusted, thereby optimizing the degree of aggressiveness of the retransmission strategy. This step realizes long-term performance balance through closed-loop control, and avoids resource waste caused by static setting.

[0094] Based on the above detailed description of the WAPI-based and reinforcement learning-based congestion control framework, the ECN-based shared bottleneck detection and packet scheduling method, and the WAPI protocol-based adaptive retransmission control method for the power Internet of Things, the implementation steps of the comprehensive congestion control and adaptive transmission optimization method based on WAPI are as follows:

[0095] Step 1, WAPI environment initialization and security association establishment. All power Internet of Things terminals and access controllers need to complete WAPI certificate authentication and key negotiation to ensure node legitimacy and transmission security. Subsequent control frames and data frames are encrypted based on the negotiated key, and the AC maintains the WAPI state information of the terminal to provide a trusted identity basis for transmission optimization, so that the whole process is protected under the WAPI security framework.

[0096] Step 2, on the basis of security initialization, deploy the congestion control module based on symbolic distillation (SymDistill-CC). This module collects decision trajectories in multiple typical scenarios in a simulation platform that simulates WAPI network characteristics, forming a dataset containing WAPI statistics (such as SNR, FLR, QL, QoS level, etc.). Through symbolic distillation, the initial strategy is converted into an interpretable symbolic decision tree, WAPI-specific operators are introduced, and genetic programming optimization rules are used to generate lightweight control strategies that adapt to different WAPI network contexts, while maintaining high performance and reducing computational overhead.​​​​

[0097] Step 3, then start the ECN and forward prediction-based packet scheduling module (ECN-FPS), solve the shared bottleneck and out-of-order problem in WAPI multi-path transmission. The module configures ECN queue threshold according to WAPI QoS level, and the receiving end feeds back congestion mark (ECE) through WAPI encrypted channel. The sending end detects the shared bottleneck sub-flow based on ECE signal, predicts the transmission amount of each sub-flow within half RTT combined with WAPI link RTT data, dynamically adjusts the sending position of slow sub-flow, avoids out-of-order from the source end, and monitors and corrects the prediction parameters in real time through WAPI link.

[0098] Step 4, finally run the WAPI security-aware adaptive retransmission module. AC obtains encrypted network parameters in real time through WAPI security channel, calculates the current network load and link utilization. The retransmission decision is based on dynamic threshold: when the load is low, retransmit the lost packets to the terminal through WAPI unicast security channel; when the load is high, select the unicast or multicast mode with smaller expected load according to the number of lost packet terminals, and all retransmission data are encrypted through WAPI. The decision parameters are dynamically adjusted according to the WAPI network state, and the balance between security and resource efficiency is realized.

[0099] Finally, it should be pointed out that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for integrated congestion control and adaptive transmission optimization based on WAPI, characterized in that: Comprise the following steps: Step 1, WAPI environment initialization and security association establishment is carried out; Step 2, on the basis of security initialization, deploy the congestion control module based on symbolic distillation, which is deployed in the simulation platform simulating the characteristics of WAPI network, collects the decision trajectory under multiple typical scenarios, forms the data set containing WAPI statistics, and converts the initial strategy into an interpretable symbolic strategy through symbolic distillation, introduces WAPI special operators, and adopts genetic programming optimization rules to generate lightweight control strategies suitable for different WAPI network contexts; Step 3, start the data packet scheduling module based on ECN and forward prediction, which configures ECN queue threshold according to WAPI QoS level, receives end feedback congestion mark ECE through WAPI encryption channel, and sends end detects shared bottleneck subflow based on ECE signal, combines WAPI link RTT data to predict the transmission amount of each subflow within half RTT, dynamically adjusts the sending position of slow subflow, and modifies the prediction parameters in real time through WAPI link monitoring; Step 4, run the WAPI security-aware adaptive retransmission module, AC obtains encrypted network parameters in real time through WAPI security channel, calculates the current network load and link utilization, and makes retransmission decision based on dynamic threshold: when the load is low, WAPI unicast security channel is used to retransmit to the terminal with packet loss; When the load is high, select the unicast or multicast mode with smaller expected load according to the number of packet loss terminals, and all retransmission data are encrypted through WAPI.

2. The method according to claim 1, wherein the method is characterized in that: The congestion control module based on symbolic distillation integrates the congestion control framework based on WAPI and reinforcement learning, extracts symbolic strategy by using teacher-student distillation mechanism, and integrates multiple context branch agents to adapt to different network conditions, realizes long-term congestion planning and dynamic rate adjustment, and the initial strategy in step 2 is the original congestion control strategy in the form of deep neural network.

3. The method according to claim 2, wherein the method further comprises: In the congestion control framework based on WAPI and reinforcement learning, WAPI branch reinforcement learning framework is also set, which first divides WAPI network conditions into three mutually exclusive contexts, then trains special branch agents for each context, and the training data comes from the teacher trajectory subset of the corresponding scene; Finally, the return data of each scene is analyzed by K-means clustering to automatically determine the branch boundary conditions.

4. The method according to claim 3, wherein the method further comprises: The PPO algorithm is used for training a teacher intelligent agent in a WAPI scene, a cross entropy loss is used as a feedback signal, rule mutation, cross and pruning are realized, an optimal symbol strategy is screened, and the cross entropy loss of the symbol strategy The formula is defined as: ； In the formula: is the teacher decision label, indicates that the rate needs to be raised, indicates that the rate needs to be reduced / maintained, is the student symbol strategy pair decision output probability, calculated by symbol rule combination, is the WAPI vector statistical symbol, is the total number of trajectory samples in the data set.

5. The method of claim 1, wherein the method further comprises: In the data packet scheduling module based on ECN and forward prediction, the computer program integrating network congestion awareness and forward prediction is integrated, and the implementation steps are as follows: Step 11: first, determine which data subflows are sharing the same physical bottleneck link, and generate a shared bottleneck candidate set; Step 12: analyze the correlation of congestion event time series of subflows in the candidate set to confirm that they indeed share the same physical bottleneck link, and form the final shared bottleneck subflow set; Step 13: For each fast sub-flow that is slower than the current slow sub-flow short fast sub-flow when the fast sub-flow belongs to one of the shared bottleneck candidate set, needs one RTT time congestion window plus 1, at this time the calculated number of data packets that can be sent by each fast sub-flow in the half RTT window is as follows: ； In the formula: representing a substream the congestion window of the substream, ; When the flow of The behavior of the flows that do not belong to any of the shared bottleneck candidate sets is similar to the traditional TCP, each RTT window congestion window is incremented by 1, at this point the number of packets that can be sent in half the RTT window is calculated for each flow The number of packets that can be sent in half the RTT window is calculated as follows: ； Step 14: Combine all the slower sub-streams The predicted packet counts of the faster substreams are summed to obtain the total; Step 15: all data packets to be sent are regarded as a logically ordered shared sending buffer.

6. The method of claim 1, wherein the method further comprises: The WAPI security-aware adaptive retransmission module integrates a power internet of things adaptive retransmission control method based on the WAPI protocol, combines the WAPI security mechanism and the dynamic retransmission strategy, and adaptively selects a unicast or multicast retransmission mode through real-time sensing of a network state.

7. The method of claim 6, wherein the method further comprises: The calculation formula of the current network load and the link utilization rate in step 4 is as follows: ； In the formula: is the average network load in the time interval is the average link utilization in the time interval , respectively represent the number of multicast and unicast retransmissions that have occurred up to time is the link bandwidth, is the multicast flow rate, is the packet length, and N1 is a power internet of things terminal in the coverage range of the base station.​​​ 8. The method of claim 7, wherein the method further comprises: In the retransmission decision process, firstly check whether the current network load exceeds the set threshold, if it satisfies the following formula, prefer to use unicast retransmission, otherwise, according to the current number of packet loss terminals , respectively calculate the expected network load under unicast and multicast mode: ； In the formula: is a dynamic adjustment parameter; The calculation formula of the expected network load in the unicast and multicast modes is as follows: ； wherein: denotes the expected normalized load of the network when unicast retransmission mode is adopted under the condition that there are packet loss terminals, denotes the expected normalized load of the network when multicast retransmission mode is adopted under the condition that there are packet loss terminals, denotes the expected number of retransmissions required for each data packet to be successfully transmitted in unicast mode, denotes the expected number of retransmission rounds required to ensure that all packet loss terminals successfully receive the data in multicast mode. By comparing and Both sizes, select the retransmission mode with the smaller expected load.

9. The method of claim 3, wherein the method further comprises: if the WAPI authentication is successful, sending a WAPI authentication success message to the WAP server; and if the WAPI authentication is not successful, sending a WAPI authentication failure message to the WAP server. By the K-means clustering process, the context center distance The criterion formula is defined as: ； wherein: is the , class context, is the is the feature, is the is the feature, is the feature weight.

10. A computer apparatus comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, causes the processor to perform the method of any one of claims 1 to 8. The processor executes the computer program to implement the steps of the method in any one of claims 1-9.

Images