Intelligent network congestion control method based on frequency domain feature analysis and dynamic weight adjustment

Abstract

The invention discloses an intelligent network congestion control method based on frequency domain feature analysis and dynamic weight adjustment, which comprises the following steps: acquiring network round-trip time RTT and queue length, and calculating a jitter value; dFT conversion is performed on the jitter value into a frequency domain, and then a high-frequency energy ratio is calculated. And constructing a state vector, and performing real-time network state estimation on the state vector based on a Kalman filtering model to obtain a state estimation value of the network. According to the high-frequency energy ratio, the weight of the RTT, the queue length and the frequency domain feature is dynamically adjusted, and then the parameters of the PID controller are adaptively adjusted according to the weight. And finally, a PID control signal generated by the PID controller is converted into a sending rate adjustment action or an explicit congestion notification, so that network congestion is inhibited in real time. According to the method, the performance of the data center network in a high-load and complex network environment can be remarkably improved, the delay is reduced, the throughput is improved, data packet loss is avoided, and the self-adaptive capability of the network is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of computer network communication, and particularly relates to an intelligent network congestion control method based on frequency domain feature analysis and dynamic weight adjustment. Background Art

[0002] With the rapid development of cloud computing, big data, Internet of Things, and 5G communication technologies, the traffic scale faced by data center networks is increasing day by day, and the complexity and dynamics of the network have also increased significantly. The transmission requirements of a large number of real-time data flows have made the network congestion problem increasingly prominent, seriously restricting the network service quality (QoS) and reducing the user experience and network performance. Therefore, how to monitor and alleviate network congestion in a timely and accurate manner has become one of the important challenges in the current data communication field.

[0003] Traditional data center network congestion control methods mainly rely on single or simple state monitoring metrics (such as simple RTT or queue length monitoring) and adopt control strategies with fixed parameters, making it difficult to achieve efficient and real-time congestion control responses in a rapidly changing network environment. Such a fixed control mechanism is prone to phenomena such as slow response, overreaction, or insufficient control when sudden fluctuations or instantaneous congestion occur in network load, resulting in serious network performance fluctuations, such as a sharp increase in delay, a significant decrease in throughput, or even a large number of packet losses.

[0004] In recent years, the Kalman Filter algorithm has been widely applied in the field of network state estimation, which can effectively estimate and predict state changes such as network delay and queue length in real time through a recursive method. However, the application of traditional Kalman Filter in the field of network congestion control is mostly limited to time-domain data and lacks effective consideration of the frequency domain characteristics of network traffic changes. Therefore, it is difficult to accurately identify and respond to short-term network jitters and high-frequency congestion events.

[0005] On the other hand, Fourier transform (especially discrete Fourier transform, DFT) technology can effectively capture the frequency domain characteristics of network state changes, especially having an obvious identification advantage for short-term and high-frequency network fluctuations. However, in the prior art, Fourier transform is usually only used for spectrum analysis and feature identification of network traffic, and thus fails to fully utilize its potential in network real-time control and congestion alleviation.

[0006] In summary, the current data center network congestion control methods still have the following main problems: First, there is a lack of a refined real-time prediction and response mechanism for network status; second, the integration of frequency domain features and time domain control methods is insufficient; third, the control parameters are fixed and cannot adaptively cope with the dynamically changing network environment. These deficiencies seriously restrict the control performance of the data center network in complex and changing real-world application scenarios, and there is an urgent need for an innovative solution that can more efficiently and accurately respond to network status changes in real time and comprehensively improve network performance and stability. Summary of the Invention

[0007] To solve the problems of insufficient real-time response, low accuracy, and fixed control strategies of current traditional network congestion control methods in complex network environments, the present invention provides an intelligent network congestion control method based on frequency domain feature analysis and dynamic weight adjustment, which can dynamically optimize control strategies by introducing frequency domain feature analysis and adaptive state estimation, accurately respond to various network environment changes, and thus improve network performance.

[0008] To achieve the above technical objectives, the technical solution of the present invention is as follows:

[0009] An intelligent network congestion control method based on frequency domain feature analysis and dynamic weight adjustment, comprising the following steps:

[0010] Step 1: Collect network round-trip time (RTT) and queue length data based on time, and then calculate the jitter value after segmenting the data in the time domain according to a fixed window size.

[0011] Step 2: Perform discrete Fourier transform (DFT) on the jitter value of each segment of RTT and queue length data respectively to convert it into the frequency domain, and then calculate the high-frequency energy ratio in the frequency domain.

[0012] Step 3: Construct a state vector including RTT deviation, queue length, and high-frequency energy ratio, and then perform real-time network state estimation on the state vector based on the Kalman filter model to obtain the RTT estimation value, queue length estimation value, and frequency domain feature estimation value of the network.

[0013] Step 4: Dynamically adjust the weights of RTT, queue length, and frequency domain features through the Sigmoid function according to the high-frequency energy ratio, and then adaptively adjust the parameters of the PID controller according to the weights.

[0014] Step 5: Convert the PID control signal generated by the PID controller into a transmission rate adjustment action or an explicit congestion notification (ECN) to suppress network congestion in real time.

[0015] Further, in the above method, Step 1 includes:

[0016] Collect the RTT and queue length data in the form of a time series to obtain the specific values of RTT and queue length corresponding to each time point;

[0017] Segment the data by a fixed time length, i.e., a window, and then calculate the differences between adjacent data in each segment, and use the calculated differences as the jitter values of this segment of data.

[0018] Furthermore, in the method, step 1 further includes the step of filtering out outliers for the jitter values of each segment of data:

[0019] Calculate the mean value of the data in each segment, then calculate the standard deviation based on the mean value, and then determine the threshold of the jitter value of this segment of data according to the standard deviation, and judge the jitter value greater than the threshold as an outlier and delete it.

[0020] Furthermore, in the method, step 2 includes:

[0021] Perform a discrete Fourier transform (DFT) on the jitter value of each segment of data according to the following formula:

[0022] ;

[0023] where is the complex frequency domain coefficient corresponding to the k-th point of the spectrum, is the length of the fixed window, represents the jitter value of each segment of data, is the natural base, represents the imaginary unit, represents pi, represents the -th sampling moment in each segment of data;

[0024] Then calculate the proportion of high-frequency energy in the frequency domain according to the following formula:

[0025] ;

[0026] where is the set frequency threshold, which determines the starting frequency of the high-frequency part, represents the energy at the k-th frequency point, is the proportion of the high-frequency component energy relative to the total energy of the signal, i.e., the proportion of high-frequency energy.

[0027] Furthermore, in the method, in step 3, the state vector is expressed as:

[0028] ;

[0029] where, represents the state vector at moment, denotes the network latency deviation at moment; denotes the queue length at moment; denotes the proportion of high-frequency energy in the frequency-domain feature at moment;

[0030] Then, the state of the network is regarded as a linear dynamic system and is represented by the following state evolution equation and observation equation:

[0031] ;

[0032] ;

[0033] where F is the state transition matrix, denotes the state vector at the previous moment of moment, B is the control input matrix, denotes the control input at the previous moment of moment,

[0034] denotes the actual observed value of RTT or queue length at is the observation matrix, denotes the observation noise.

[0035] Furthermore, in the state evolution equation of the above method, the state transition matrix F is expressed as:

[0036] ;

[0037] where denotes the influence coefficient of queue length on RTT deviation, denotes the influence coefficient of RTT deviation on the proportion of high-frequency energy, denotes the influence coefficient of RTT deviation on queue length, denotes the relationship coefficient between network load and short-term fluctuation, denotes the influence coefficient of the proportion of high-frequency energy on RTT deviation, denotes the influence coefficient of the proportion of high-frequency energy on queue length;

[0038] The control input matrix is expressed as:

[0039] ;

[0040] where Indicates the influence coefficient rate of the control input on the RTT deviation, and the influence coefficient of the rate on the RTT deviation, Indicates the adjustment coefficient of the rate on the queue length, Indicates the suppression coefficient of the rate on the high-frequency energy, Indicates the constraint coefficient of the rate on the queue growth, Indicates the coupling coefficient of the rate on the frequency-domain characteristics, Indicates the smoothing coefficient of the rate on the delay fluctuation;

[0041] In the observation equation, the observation matrix Is expressed as:

[0042] ;

[0043] Where , And Respectively indicate the influence coefficients of the RTT deviation, queue length, and high-frequency energy occupancy ratio on the RTT measurement value; , And Respectively indicate the influence coefficients of the RTT deviation, queue length, and high-frequency energy occupancy ratio on the queue length measurement value.

[0044] Furthermore, in the method described above, in step 3, the real-time network state estimation of the state vector based on the Kalman filter model includes the following steps:

[0045] Perform real-time network state estimation through two stages of prediction and correction:

[0046] In the prediction stage, it includes the predicted state and predicted error covariance matrix obtained through the following formula:

[0047] ;

[0048] ;

[0049] Where Indicates the predicted value of the state at time, Is the optimal state estimation value at the previous moment of time;

[0050] Indicates the error covariance matrix predicted at time, Indicates The error covariance matrix predicted at the previous moment of time, The superscript of Indicates matrix transpose, Indicates the covariance matrix of the process noise;

[0051] In the calibration phase, it includes first calculating the Kalman gain through the following formula, and then combining the observed data and the predicted state based on the Kalman gain to obtain the optimal state estimate value:

[0052] ;

[0053] ;

[0054] where represents the Kalman gain, The superscript of represents matrix transpose, represents the covariance matrix of the observation noise;

[0055] represents the optimal state estimate value of the network state at time, is the innovation quantity, representing the difference between the observed value and the predicted value;

[0056] Then, the error covariance matrix is updated through the following formula:

[0057] ;

[0058] where represents the updated error covariance matrix and is used as the at the next moment of time k, represents the identity matrix.

[0059] Furthermore, in the method described above, in step 4, the process of dynamically adjusting the weights of RTT, queue length, and frequency domain characteristics through the Sigmoid function includes:

[0060] First, use the Sigmoid function to dynamically map the high-frequency energy ratio to the weight adjustment coefficient, thereby dynamically adjusting the RTT weight, and then further set the queue length weight and high-frequency energy ratio weight based on the RTT weight:

[0061] ;

[0062] ;

[0063] ;

[0064] where represents the weight of RTT, is the adjustment sensitivity coefficient, is the threshold of the high-frequency energy ratio, is the natural base;

[0065] Indicates the queue length weight;

[0066] Indicates the weight of the high-frequency energy ratio.

[0067] Furthermore, in the method, in step 4, adaptively adjusting the parameters of the PID controller according to the weights includes:

[0068] Adjusting the proportional gain of the PID controller based on the following formulas , integral gain and derivative gain respectively for the three parameters:

[0069] ;

[0070] ;

[0071] ;

[0072] where is the initial proportional gain, is the adjustment factor of the proportional gain;

[0073] is the initial integral gain, is the adjustment factor of the integral gain, represents the control error of the PID controller, is the error threshold, represents other situations;

[0074] is the initial derivative gain, is the adjustment factor of the derivative gain.

[0075] Furthermore, in the method, step 5 includes:

[0076] First, calculate the control error based on the network estimated state :

[0077] ;

[0078] where is the current real-time estimated round-trip delay, and are the target network delay and the queue length at time respectively, both are included in is the target queue length;

[0079] Then, based on to calculate the control output of the PID controller :

[0080] ;

[0081] where represents the mixed error value at the historical moment, represents the control period, represents the mixed error value at the previous moment of the

[0082] Then, according to to adjust the transmission rate :

[0083] ;

[0084] ;

[0085] where represents the adjustment value of the transmission rate, is the conversion factor that maps the PID control signal to the actual network control quantity, represents the transmission rate at the next moment of the represents the transmission rate at the

[0086] Meanwhile, the transmission rate is monitored in real time. When the transmission rate drops to the preset threshold , an ECN marking operation is triggered to mark the ECN field in the IP header of the data packet, so that the source node actively reduces the transmission rate after receiving the data packet with the ECN mark, thereby completing the explicit congestion notification ECN.

[0087] The technical effect of the present invention is that the present invention combines the frequency domain features with the Kalman filter, proposes a dynamic weight allocation method based on the high-frequency energy ratio, and combines the adaptive PID parameter adjustment to achieve precise and real-time control of the network state. By analyzing the frequency domain features to capture the instantaneous fluctuations and short-term changes in the network, the system can dynamically adjust the control strategy according to the real-time network state, making the network control response more sensitive and accurate. The present invention can significantly improve the performance of the data center network in high-load and complex network environments, reduce latency, increase throughput, avoid packet loss, and enhance the adaptive ability of the network. By real-time estimating the network state and dynamically adjusting the PID controller parameters, the system can accurately respond to various changes in the network state and ensure the stability and efficiency of the network performance in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 Schematic diagram of the functions of the steps in the embodiments of the present invention;

[0089] Figure 2 Schematic diagram of the corresponding frequency-domain energy spectrum of the RTT jitter time-domain signal in the embodiments of the present invention, where (a) is the RTT jitter time-domain signal diagram and (b) is the corresponding frequency-domain energy spectrum diagram.

[0090] Figure 3 Schematic diagram of the corresponding frequency-domain energy spectrum of the queue length jitter time-domain signal in the embodiments of the present invention, where (a) is the queue length jitter time-domain signal diagram and (b) is the corresponding frequency-domain energy spectrum diagram.

[0091] Figure 4 Schematic flowchart of step 5 in the embodiments of the present invention. Detailed implementation manners

[0092] Refer to Figure 1 , an intelligent network congestion control method based on frequency-domain feature analysis and dynamic weight adjustment provided in this embodiment includes the following steps:

[0093] Step 1: Collect network round-trip time (RTT) and queue length data based on time, and then calculate the jitter value after segmenting the data in the time domain according to a fixed window size.

[0094] In this step, the time-series data of RTT and queue length are first obtained in real time from the network. Specifically, the first signals collected are the change data of RTT and queue length. The data collection will be stored in the form of a time series, that is, each data point corresponds to a measurement value of a time stamp For example, for the RTT data, after storing it in the time series, a sequence is obtained:

[0095] ;

[0096] wherein the corresponds to the time stamp and the specific network round-trip time value.

[0097] For the queue length data, a similar sequence is obtained in the same way:

[0098] ;

[0099] Assume that the RTT corresponding to each sampling point is , then the RTT jitter is:

[0100] ;

[0101] Collect queue length data from network nodes (such as switches or routers) and calculate the fluctuations in the queue length. Assume that the queue length at each sampling moment is , then the queue length jitter is:

[0102] .

[0103] Then segment the collected time-series data into fixed-time-length windows. In the following steps, the data within each window will be separately subjected to Fourier transform, which helps analyze the frequency-domain characteristics of the signal at different time periods.

[0104] Next, smooth the RTT and queue length data, that is, calculate the mean and standard deviation to filter out outliers. Assume that the data window size is , and the calculation of the mean and standard deviation for the RTT data is:

[0105] ;

[0106] ;

[0107] Similarly, calculate the mean and standard deviation for the queue length:

[0108] ;

[0109] ;

[0110] Based on the standard deviation, outliers can be filtered out:

[0111] ;

[0112] ;

[0113] For or that satisfy the above inequalities are regarded as outliers and removed from the current data window. This method can effectively suppress the interference of abnormally high jitter on the extraction of frequency-domain features.

[0114] Step 2: Perform the discrete Fourier transform (DFT) on the jitter values of each segment of the RTT and queue length data to convert them to the frequency domain, and then calculate the proportion of high-frequency energy in the frequency domain.

[0115] To extract the frequency-domain characteristics of the RTT / queue jitter, in this embodiment, the discrete Fourier transform (DFT) is used to convert the time-domain signal to the frequency domain.

[0116] For the RTT data, its jitter sequence can be constructed as:

[0117] ;

[0118] For the queue length data, its jitter sequence can be constructed as follows:

[0119] ;

[0120] where represents the end-to-end delay value (original RTT data) obtained at the th sampling moment, represents the discrete sampling time point index (integer), represents the time-domain jitter sequence (differential result) calculated based on RTT.

[0121] According to the collected jitter time sequence, its DFT can be expressed as:

[0122] ;

[0123] where is the complex frequency domain coefficient corresponding to the kth point of the spectrum, is the length of the fixed window, represents the jitter value of each segment of data, that is, the jitter sequence of RTT or queue length obtained above. is the natural base, represents the imaginary unit, represents pi, represents the th sampling moment in each segment of data.

[0124] Through the spectrum the energy of the signal in different frequency bands can be calculated. Through Fourier transform, the jitter signal in the time domain is thus converted into the frequency domain, obtaining the amplitude and phase information of each frequency component, and further identifying the frequency characteristics of the signal.

[0125] After Fourier transform, through the spectrum the energy of the signal in different frequency bands can be further calculated. To measure the proportion of high-frequency fluctuations in the signal, this embodiment proposes the definition of the high-frequency energy ratio as follows

[0126] ;

[0127] where represents the actual frequency value corresponding to this frequency component, is the sampling frequency. is the set frequency threshold, which determines the starting frequency of the high-frequency part. Generally, this value can be set according to the typical periodic characteristics of network traffic. represents the energy at the kth frequency point.

[0128] It is the ratio of the high - frequency component energy to the total signal energy, reflecting the intensity of short - term fluctuations in the network. This index is used to characterize the intensity of short - term and sudden fluctuations in the network state. The larger the value, the more significant the high - frequency jitter and the higher the risk of instantaneous network congestion.

[0129] In this step, the output high - frequency energy ratio As a key feature, it can effectively reflect the degree of dynamic change of network jitter. If the value is high, it indicates that there are more instantaneous changes in the network, and there may be phenomena such as bursty traffic or temporary congestion; if it is low, it means that the network is more stable and may be in a relatively stable working state.

[0130] Frequency - domain features (such as the high - frequency energy ratio) are inputs for subsequent adaptive state estimation and dynamic weight control in this embodiment. Refer to Figure 2 and Figure 3 . Under different RTT jitter time - domain signals and queue - length jitter time - domain signals, there will be different frequency - domain energy spectra. Among them, Figure 2 the high - frequency energy ratio in (b) of Figure 3 is 23.07%, and

[0131] When performing adaptive state estimation, these frequency - domain features are used as an important dimension of the system state, and state estimation is carried out through Kalman filtering, combined with other network states (such as RTT, queue length, etc.) to analyze and predict network behavior.

[0132] When performing dynamic weight control, frequency - domain features (especially the high - frequency energy ratio) are used to adjust the control strategy. For example, in the case of larger high - frequency energy, a stronger control response can be selected to cope with rapid fluctuations; while in the case of a larger low - frequency energy ratio, the control strategy can be adjusted to better adapt to long - term network changes.

[0133] Since most traditional network performance monitoring methods rely on time - domain analysis, such as directly monitoring statistical information such as the average value, maximum value, and variance of RTT and queue length. This method can reflect the general state of the network, but it is difficult to capture sudden and high - frequency instantaneous fluctuations in the network (such as sudden increases in network load, traffic floods, etc.). In this embodiment, the time - domain signal is converted into a frequency - domain signal through Fourier transform, which can effectively capture these frequency characteristics, especially high - frequency noise in the network.

[0134] Fourier transform can extract frequency - domain features from time - domain signals, thereby effectively identifying instantaneous fluctuations (high - frequency noise) and long - term trends (low - frequency changes) in the network.

[0135] RTT and queue length data can reflect key performance indicators in the network. Combining with the Fourier transform, the frequency characteristics of fluctuations in the network can be captured more accurately.

[0136] Using the proportion of high-frequency energy as a feature can clearly classify high-frequency noise and low-frequency trends, which helps the network control system adopt appropriate strategies according to the actual fluctuations of the network.

[0137] Step 3: Construct a state vector containing RTT deviation, queue length, and the proportion of high-frequency energy. Then, based on the Kalman filter model, perform real-time network state estimation on the state vector to obtain the RTT estimation value, queue length estimation value, and frequency domain feature estimation value of the network.

[0138] In this step, the state vector is expressed as:

[0139] ;

[0140] where represents the state vector at time, represents the network delay deviation at time; represents the queue length at time; represents the proportion of high-frequency energy in the frequency domain feature at time. It reflects the network delay fluctuation in real time and shows the change trend of network delay. It reflects the load situation and congestion level of the network. It describes the severity of short-term fluctuations and sudden situations in the network. The frequency domain feature clearly reflects the potential risk of instantaneous network congestion.

[0141] Then, model the network state estimation as a linear dynamic system and use the Kalman filter to perform real-time estimation and prediction of the system state. The state space model includes two major equations:

[0142] State evolution equation (time update):

[0143] ;

[0144] where F is the state transition matrix, represents the state vector at the previous moment of time, B is the control input matrix, represents the control input at the previous moment of The process noise at the previous moment of the current moment, assumed to be zero-mean Gaussian noise with covariance Q, is used to describe the uncertainty in the change of network state. For example, sudden traffic or equipment failures may cause significant fluctuations in RTT or queue length, which cannot be directly predicted by the model, so process noise is introduced. .

[0145] In the state evolution equation, the state transition matrix F represents the change of network state from moment to moment, which is used to describe the natural change trend of network state under uncontrolled conditions. Its elements can be fitted based on simulation data or set empirically, reflecting the coupling and inertia among RTT jitter, queue length, and frequency-domain energy.

[0146] The state transition matrix F is expressed as:

[0147] ;

[0148] where : The influence coefficient of queue length on RTT deviation. Since network congestion usually increases the queue length, which in turn leads to an increase in RTT, this term is positive. The value depends on the network topology, traffic characteristics, and control strategy. Usually, this value can be estimated through experiments or simulation data. Suppose experimental data shows that for every 10-packet increase in queue length, the RTT deviation increases by approximately 20 ms, then can be expressed as 0.02 (i.e., for every one-unit increase in queue length, the RTT deviation increases by 0.02).

[0149] : The influence coefficient of RTT deviation on the proportion of high-frequency energy. The change of RTT is closely related to the instantaneous fluctuations and frequency-domain characteristics (high-frequency components) in the network. The proportion of high-frequency energy is usually used to measure the short-term fluctuations in the network. When the high-frequency components are strong, the RTT deviation may fluctuate significantly. It can be assumed that there is a certain positive correlation between the RTT deviation and the proportion of high-frequency energy. If the high-frequency energy increases by 10%, the RTT deviation increases by 5 ms, then may take a value of 0.05.

[0150] : The influence coefficient of RTT deviation on queue length. Generally, an increase in network load (increase in RTT) will lead to an increase in queue length, so this term is positive. According to the actual network load situation, if the RTT increases by 10 ms, the queue length may increase by 5 data packets, then may take a value of 0.5.

[0151] : An increase in queue length usually means an increase in network load, which may lead to more severe short-term fluctuations in the network, thus affecting the high-frequency energy in the frequency domain. That is, when the queue length is large, there may be more instantaneous changes in the network. reflects the correlation coefficient between network load and short-term fluctuations. Usually, should be a positive value, indicating that an increase in queue length is usually accompanied by an increase in the proportion of high-frequency energy. If the queue length increases by 10 packets and the proportion of high-frequency energy in the network increases by 5%, then may take the value of 0.05.

[0152] : The influence coefficient of the proportion of high-frequency energy on the RTT deviation. Although the relationship between high-frequency energy and RTT is not as direct as that of queue length, there may still be a certain feedback effect. This item can be determined according to actual data. may be small because the impact of high-frequency fluctuations on latency is not as direct as that of queue length. In some cases, can be taken as a small positive value, such as 0.02, indicating that the influence of high-frequency energy on RTT is weak.

[0153] : The influence coefficient of the proportion of high-frequency energy on queue length. Short-term load fluctuations are usually accompanied by changes in queue length, so this item is positive. represents the influence of high-frequency energy on queue length. Usually, should be a positive value, indicating that high-frequency fluctuations may lead to an increase in queue length. If the high-frequency energy increases by 10% and the queue length increases by 5 packets, then may take the value of 0.5.

[0154] Control input matrix , describing the influence of control input on the state, represents the modulation effect of control input (such as transmission rate adjustment) on the system state. Usually, it mainly acts on the queue state and indirectly affects the latency and jitter characteristics.

[0155] Control input matrix is expressed as:

[0156] ;

[0157] Among them, : The influence coefficient of rate on RTT deviation.

[0158] Control input (transmission rate adjustment) directly affects the deviation change of network transmission delay. Increasing the transmission rate can reduce the degree of link congestion, thereby reducing the amplitude of RTT fluctuations. Usually, It should be a negative value, indicating an inverse relationship between the rate increase and the RTT deviation. According to the experimental data: when the sending rate is increased by 10 Gbps, the RTT deviation is reduced by an average of 2 ms, then it may take a value of -0.02. In a high-speed network environment, when the rate approaches the bandwidth limit, the adjustment effect of this coefficient is enhanced by about 30%.

[0159] : The adjustment coefficient of the rate on the queue length.

[0160] Quantify the adjustment effect of the sending rate on the switch buffer load. Reasonably increasing the sending rate can accelerate queue emptying, but excessive increase may lead to instantaneous congestion. It reflects the control efficiency of the rate adjustment on the queue length. Normally, this coefficient is positive. The measured data shows that: in a 25 Gbps link, when the sending rate increases by 1 Gbps, the queue length is reduced by an average of 8 packets (about a 12% reduction), then it may take a value of 0.08. When the queue utilization rate exceeds 85%, the adjustment effect is enhanced by 50%.

[0161] : The suppression coefficient of the rate on the high-frequency energy.

[0162] The regulation effect of the sending rate increase on the high-frequency components in the network spectrum characteristics. Appropriately increasing the sending rate can smooth network traffic, reduce burst fluctuations, and thus reduce the proportion of high-frequency energy. This coefficient should be negative, indicating an inverse relationship between the rate increase and the high-frequency energy. The experimental statistics show that: when the sending rate increases by 1 Gbps, the proportion of high-frequency energy is reduced by an average of 0.15%, then it may take a value of -0.15. In the network high-load (>70% bandwidth occupancy) scenario, the suppression effect is increased by 40%.

[0163] : The constraint coefficient of the rate on the queue growth.

[0164] The control ability of the sending rate adjustment on the buffer queue expansion. Reasonably increasing the rate can inhibit the excessive growth of the queue and prevent buffer overflow. This coefficient should be negative, indicating that the rate increase can effectively constrain the queue growth. The verification data shows that: when the rate increases by 1 Gbps, the queue growth rate is reduced by an average of 7%, then it may take a value of -0.07. When the queue is close to full load (>90% capacity) critical state, the constraint efficiency is increased to 3 times the benchmark value.

[0165] : The coupling coefficient of the rate on the frequency domain characteristics.

[0166] The dynamic correlation effect between transmission rate adjustment and changes in the network frequency domain characteristics. Although rate increase can alleviate congestion in the main frequency band, it will indirectly affect the high-frequency energy distribution through changes in traffic characteristics. This coefficient is usually positive, indicating that an increase in rate will slightly increase the high-frequency energy. The measured results show that for every 1 Gbps increase in the transmission rate, the average proportion of high-frequency energy increases by 0.03%, then may take a value of 0.03. This coupling effect needs to be jointly optimized, and the compensation efficiency in a typical network environment can reach 92%.

[0167] : The smoothing coefficient of rate on delay fluctuation.

[0168] The stabilizing effect of transmission rate optimization on transmission delay jitter. Reasonably adjusting the transmission rate can suppress the irregular fluctuations in network transmission, thereby smoothing the RTT changes. This coefficient should be negative, indicating an inverse relationship between rate increase and delay fluctuation. Experimental analysis shows that for every 1 Gbps increase in the transmission rate, the standard deviation of RTT decreases by an average of 0.04 ms, then may take a value of -0.04. In the scenario of bursty traffic, the smoothing effect of this coefficient can be improved by 60%; when the queue length reaches more than 50% of the capacity, the smoothing efficiency increases linearly.

[0169] The state space model also includes an observation equation (measurement update). The main task of the observation equation is to relate the internal state of the system (RTT deviation, queue length, and proportion of high-frequency energy) to measurable network parameters (such as the actually measured RTT and queue length). This process effectively reduces the prediction error and enables the system to better adapt to environmental changes. The observation equation is expressed as:

[0170] ;

[0171] where represents the actual observed value of RTT or queue length at time is the observation matrix, represents the observation noise.

[0172] The observation matrix is expressed as:

[0173] ;

[0174] where : The influence coefficient of RTT deviation on the RTT measurement value.

[0175] The change in the RTT deviation directly affects the measurement result of the end-to-end delay. When network congestion intensifies, the increase in the RTT deviation will significantly increase the RTT measurement value. Usually, should be positive, indicating that an increase in the RTT deviation usually leads to an increase in the RTT measurement value. If the RTT measurement value increases by 4.9 ms for every 5 ms increase in the RTT deviation, then may take a value of 0.98.

[0176] : The influence coefficient of the queue length on the RTT measurement value.

[0177] The growth of the queue length indirectly affects the RTT measurement through the buffer delay effect. When the switch buffer accumulates, the queuing delay of the data packet increases, resulting in an increase in the RTT measurement value. Usually, should be positive, indicating that an increase in the queue length is accompanied by an increase in the RTT measurement value. If the RTT measurement value increases by 0.5 ms for every 10 additional data packets in the queue length, then may take a value of 0.05.

[0178] : The influence coefficient of the high-frequency energy occupancy ratio on the RTT measurement value.

[0179] The high-frequency energy fluctuation reflects the network burst traffic and acts on the delay measurement through the congestion control mechanism. Short-term traffic bursts will trigger a rapid congestion response, which may inhibit the growth of the RTT. Usually, should be negative, indicating that an increase in the high-frequency energy will inhibit the RTT fluctuation amplitude. If the RTT measurement value decreases by 0.3 ms for every 10% increase in the high-frequency energy, then may take a value of -0.03.

[0180] : The influence coefficient of the RTT deviation on the queue measurement value.

[0181] The increase in the RTT deviation indicates an increase in the network load and is indirectly reflected in the queue accumulation trend. The increase in the transmission delay will cause the data packet to stay in the buffer for a longer time. Usually, should be positive, indicating that an increase in the RTT deviation is often accompanied by an increase in the queue length. If the queue measurement value increases by 0.6 data packets for every 5 ms increase in the RTT deviation, then may take a value of 0.12.

[0182] : The mapping coefficient of the queue state to the queue measurement value.

[0183] The actual state of the switch buffer directly determines the queue measurement value. This coefficient includes correction factors such as the measurement device error and the system response delay. Usually, Slightly greater than 1.0, indicating that the queue measurement slightly amplifies the true state. If the queue state is 100 data packets, and the measurement shows 102 packets, then it may take a value of 1.02.

[0184] : The correlation coefficient of the high-frequency energy proportion to the queue measurement.

[0185] The traffic burst caused by the sudden increase in high-frequency energy will trigger an instantaneous accumulation in the power grid buffer. The short-term impact of burst packets may cause instantaneous fluctuations in the queue depth. Generally, it should be a positive value, indicating that an increase in high-frequency energy is accompanied by a short-term increase in the queue measurement. If the high-frequency energy increases by 15% and the queue measurement temporarily increases by 1.2 data packets, then it may take a value of 0.08.

[0186] The observation matrix is used to describe how the state vector is mapped to the observation space. For network state estimation, the observation matrix can reflect the influence of different states on the observed quantity. For example, assume the observed quantities include: the actually measured RTT ( ) and the actually measured queue length ( ). In this case, the observation equation can be written as:

[0187] ;

[0188] In this step, real-time network state estimation is carried out based on the Kalman filter model through two stages: prediction and correction:

[0189] 1. Prediction (time update).

[0190] In the prediction stage, the state and error covariance at the current moment are predicted based on the dynamic model of the system (i.e., the state transition equation). This process does not rely on the observed data, but only makes predictions based on the state estimation and control input at the previous moment of the system.

[0191] Predicted state:

[0192] ;

[0193] Among them, is the predicted value of the state at the moment predicted according to the state estimation and control input at the previous moment, that is, the prior estimation of the state through the system model before obtaining the current observed data. is the optimal state estimation value at the previous moment of the moment.

[0194] Predicted error covariance matrix:

[0195] ;

[0196] Among them, represents the error covariance matrix of the prediction at time, represents the error covariance matrix of the prediction at the previous time of The superscript of represents matrix transpose, represents the covariance matrix of the process noise.

[0197] 2. Correction (measurement update).

[0198] The goal of the correction stage is to correct the prediction result by fusing the observation data. The Kalman filter adjusts the state estimate using the Kalman gain by comparing the difference between the predicted network state and the actual observation data to obtain a more accurate network state.

[0199] ;

[0200] Among them: .

[0201] The Kalman gain is used to balance the weights of the prediction uncertainty and the covariance matrix R of the observation noise, where R is the statistical description of the observation noise. If the prediction error is large and the observation noise is small, the Kalman gain will be higher, indicating that the system relies more on the observed values.

[0202] Through the Kalman gain, the Kalman filter combines the observation data and the predicted state to update the state estimate. The updated state estimate is the optimal estimate of the network state before

[0203] ;

[0204] Among them, is the innovation, which represents the difference between the observed value and the predicted value. The Kalman filter updates the state estimate by correcting the innovation.

[0205] While updating the state estimate, it is also necessary to update the error covariance matrix to reflect the influence of the observation data on the accuracy of the state estimate. The updated error covariance matrix is usually smaller, indicating that the uncertainty of the state estimate is reduced after introducing the observation data.

[0206] ;

[0207] is the updated error covariance matrix, representing the uncertainty of the updated state estimate, and is used as the at the next moment of to use.

[0208] By combining the prediction and correction steps, the Kalman filter can effectively and dynamically estimate the RTT deviation, queue length, and high-frequency energy ratio in network state estimation. Through the Kalman gain, the Kalman filter adjusts the state estimate according to the prediction error and observation noise, ensuring that the network control system can adapt to network state changes in real time and maintain high performance in an uncertain environment. This recursive estimation process enables the congestion control scheme to precisely control the network state, thereby optimizing the network's throughput, latency, and stability.

[0209] Step 4: Dynamically adjust the weights of RTT, queue length, and frequency domain features through the Sigmoid function according to the high-frequency energy ratio, and then adaptively adjust the parameters of the PID controller according to the weights.

[0210] The main function of this step is to dynamically adjust the weights in the network control strategy according to the network state estimation results output in the previous step to achieve refined congestion control and network state optimization. This step analyzes the RTT deviation, queue length, and high-frequency energy ratio in the frequency domain, comprehensively evaluates the congestion status of the network, and dynamically allocates the weights of each state variable in the control strategy based on this evaluation. This step also includes a PID parameter adaptive adjustment mechanism, which uses dynamic weight information to optimize the parameters of the PID controller in real time, thereby effectively reducing network latency, reducing queue backlogs, and improving the overall stability and performance of the network.

[0211] In this step, the weights are first dynamically adjusted, and the dynamic adjustment is based on the real-time relationship between the high-frequency energy ratio and the network state:

[0212] First, calculate the dynamic weight distribution coefficient:

[0213] Use the sigmoid function to dynamically map the high-frequency energy ratio to the weight adjustment coefficient:

[0214] ;

[0215] where represents the weight of RTT, is the base of the natural logarithm, is the coefficient for adjusting the sensitivity, controlling the steepness of the sigmoid function (for example, taking a value of 5). is the threshold of the high-frequency energy ratio. When the high-frequency energy ratio exceeds this threshold, the RTT weight will increase significantly, indicating that there are large instantaneous fluctuations in the network and it is necessary to adjust the RTT faster to reduce the delay.

[0216] Correspondingly, the queue length weight and the high-frequency energy ratio weight are set as:

[0217] ;

[0218] ;

[0219] Such settings ensure that the sum of the RTT weight and the queue length weight remains 1, while the high-frequency energy weight reaches the maximum when the two weights are multiplied, enhancing the quick response to instantaneous network congestion. When the network has severe instantaneous fluctuations ( relatively high), the weight of the RTT deviation increases to quickly relieve the instantaneous network delay; when the network is stable ( relatively low), the queue length weight increases, paying more attention to long-term congestion control and load balancing. This dynamic weight adjustment mechanism enables the network control strategy to intelligently adjust the response intensity under different congestion scenarios.

[0220] The smooth adjustment characteristic of the Sigmoid function ensures that the change of the weight is not too drastic, thus avoiding the drastic fluctuation of the control strategy when the network changes rapidly and enhancing the stability of the system. This smooth characteristic also improves the robustness of the control system and enables it to operate efficiently in different network environments.

[0221] Then, according to the result of the dynamic weight allocation, the parameters of the PID controller (Proportional-Integral-Derivative controller) ( ) are further adjusted adaptively:

[0222] Proportional gain ( ) adjustment:

[0223] ;

[0224] Among them, is the initial proportional gain. is the adjustment factor of the proportional gain. is the RTT control weight dynamically adjusted according to the high-frequency energy ratio (frequency domain feature), reflecting the severity of the RTT deviation in the network.

[0225] Proportional gain ( )(It) reflects the influence of the current error on the output control signal. For network control, when the deviation of the network delay (RTT) is large, the proportional gain needs to be adjusted rapidly to quickly respond to network fluctuations. The adaptive adjustment of the proportional gain is carried out according to the RTT control weight. As the network RTT changes, it will be adjusted according to the adjusted weight to enhance the response to delay fluctuations.

[0226] Integral gain ( ) adjustment:

[0227] Integral gain ( ) controls the cumulative effect of past errors and helps to eliminate the steady-state error of the system. Over-reliance on integral control may lead to overshoot and system oscillation. Therefore, when the network load is stable, the integral gain needs to be adjusted appropriately.

[0228] The adaptive adjustment of the integral gain is achieved through the queue length control weight. When the queue length (i.e., network load) is high, the system needs a stronger integral response to eliminate the steady-state error; while when the queue is small or the system is stable, the integral gain should be reduced appropriately.

[0229] When the network error exceeds a certain threshold, the integral term becomes effective:

[0230] ;

[0231] is the initial integral gain. is the adjustment factor. is the control weight of the queue length, reflecting the load and network stability. is an error threshold, and only when the error exceeds this threshold will the integral term take effect.

[0232] Differential gain ( ) adjustment:

[0233] Differential gain ( ) controls the rate of change of the error and is used to reduce the overshoot and oscillation of the system. In a network environment, the proportion of high-frequency energy reflects the instantaneous fluctuations in the network. The adjustment of the differential gain can help the system suppress the influence of these short-term fluctuations on the control output. The adaptive adjustment of the differential gain is achieved through the proportion of high-frequency energy. When there are large instantaneous fluctuations in the network (for example, the proportion of high-frequency energy is high), the differential gain needs to be increased to reduce the influence of short-term fluctuations on network control.

[0234] The adjustment formula for the differential gain is:

[0235] ;

[0236] is the initial differential gain, is the adjustment factor of the differential gain.

[0237] The main innovation of this step lies in dynamically adjusting the weights of each state variable in the network congestion control strategy in real time according to the frequency domain characteristics (high-frequency energy ratio). Thus, it overcomes the problem of insufficient response of traditional fixed-weight control strategies to network changes and achieves high adaptability of congestion control. In addition, the adaptive adjustment mechanism of PID parameters uses dynamic weights to optimize PID parameters, enabling the controller to adjust the control strength in real time in a complex network environment, thereby effectively reducing network latency, balancing network load, and improving the overall throughput and performance stability of the network. This highly adaptable and real-time optimization ability significantly enhances the robustness, real-time performance, and accuracy of the network congestion control system.

[0238] Step 5: Convert the PID control signal generated by the PID controller into a transmission rate adjustment action or an explicit congestion notification ECN to suppress network congestion in real time.

[0239] See Figure 4 , in this step, the PID control signal is the core data directly input to the network execution layer and is generated by the PID controller in the dynamic weight control layer. The specific calculation method of the PID control signal is as follows:

[0240] First, define the control error , which is defined based on the network estimated state as follows:

[0241] ;

[0242] where is the current real-time estimated round-trip delay (unit: milliseconds), which is calculated from the network state data through Kalman filtering, .

[0243] is the target delay expected by the network and is set according to the network service level agreement (SLA). is the current real-time estimated queue length (unit: number of data packets), which is obtained through monitoring the buffer status of the switch. is the target queue length (unit: number of data packets), which is dynamically configured according to the buffer capacity of the switch and is usually set to the queue length under the ideal network load (such as the queue length under low load). , , is the dynamic weight coefficient, which is dynamically determined by the Sigmoid function driven by the frequency domain characteristics.

[0244] The specific calculation of the control output of the non-linear PID controller is as follows:

[0245] ;

[0246] where represents the mixed error value at the historical moment, represents the control period, represents the mixed error value at the previous moment of the , , The PID parameters are adjusted in real time through an adaptive PID regulator to ensure stability and fast responsiveness when the network state changes.

[0247] In this step, the actual action of the transmission rate adjustment is mapped from the PID control output to the change in the transmission rate (Rate) of the specific network device. The actual adjustment rule of the transmission rate is:

[0248] Assume that the transmission rate of the network device at is (in Mbps), then the rate adjustment strategy for the next moment is:

[0249] ;

[0250] while is determined according to the PID control signal :

[0251] ;

[0252] where: The coefficient is a conversion factor that maps the PID control signal to the actual network control quantity (the conversion factor is generally determined according to the network characteristics and the device bandwidth range).

[0253] The negative sign indicates that when the PID output is positive, it means that the current network condition is good, the error decreases, and the network can increase the transmission rate; when the PID output is negative, it means that the network condition deteriorates, and the transmission rate needs to be reduced to relieve congestion.

[0254] For example:

[0255] If = 5, γ = 0.1, then the transmission rate decreases by 0.5 Mbps, that is, the rate becomes slower;

[0256] If = -4 and γ = 0.1, the transmission rate increases by 0.4 Mbps, i.e., the rate is improved.

[0257] In the actual control process, if it is found that the transmission rate drops to a certain threshold (such as below the set threshold , it means that the network is in a relatively serious congestion state. At this time, the network execution layer will trigger the ECN marking operation.

[0258] ECN marking trigger conditions:

[0259] ;

[0260] After the ECN marking is triggered, the network device marks the ECN field in the IP header of the data packet (usually set to 11).

[0261] After the source node receives the data packet with the ECN marking, it will actively reduce the transmission rate (such as the CWR mechanism in TCP), alleviating the network congestion situation from the source.

[0262] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code. The present application is a device that, according to the flowchart and / or block diagram of the method, device (system), and computer program product of the embodiments of the present application, is used to generate functions specified in one or more processes of the flowchart and / or one or more blocks of the block diagram by instructions executed by a processor. These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured product including an instruction device that implements the functions specified in one or more processes of the flowchart and / or one or more blocks of the block diagram. These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes of the flowchart and / or one or more blocks of the block diagram.

[0263] It should be emphasized that the examples of the present invention are illustrative rather than restrictive, and therefore the present invention is not limited to the examples in the specific embodiments. Any other embodiments derived by those skilled in the art based on the technical solution of the present invention that do not depart from the purpose and scope of the present invention, whether modified or replaced, also belong to the protection scope of the present invention.

Claims

1. An intelligent network congestion control method based on frequency domain feature analysis and dynamic weight adjustment, characterized in that The steps include: Step 1: Collect the network round-trip time (RTT) and queue length data based on time, and then segment the data in the time domain according to a fixed window size and calculate the jitter value. Step 2: Perform the discrete Fourier transform (DFT) on the jitter value of each segment of the RTT and queue length data respectively to convert it into the frequency domain, and then calculate the proportion of high-frequency energy in the frequency domain. Step 3: Construct a state vector including the RTT deviation, queue length, and proportion of high-frequency energy, and then perform real-time network state estimation on the state vector based on the Kalman filter model to obtain the RTT estimation value, queue length estimation value, and frequency domain feature estimation value of the network. Step 4: Dynamically adjust the weights of the RTT, queue length, and frequency domain features through the Sigmoid function, and then adaptively adjust the parameters of the PID controller according to the weights. Step 5: Convert the PID control signal generated by the PID controller into a transmission rate adjustment action or an explicit congestion notification (ECN) to suppress network congestion in real time.

2. The method according to claim 1, wherein The said Step 1 includes: Collect the RTT and queue length data in the form of a time series to obtain the specific values of the RTT and queue length corresponding to each time point. Segment the data according to a fixed time length, i.e., the window, and then calculate the difference between adjacent data in each segment, and use the calculated differences as the jitter value of this segment of data.

3. The method according to claim 1, characterized in that The said Step 1 further includes the step of filtering out outliers from the jitter value of each segment of data: Calculate the mean of the data in each segment, then calculate the standard deviation according to the mean, and then determine the threshold of the jitter value of this segment of data according to the standard deviation, and judge the jitter value greater than the threshold as an outlier and delete it.

4. The method according to claim 1, characterized in that, The said Step 2 includes: Perform the discrete Fourier transform (DFT) on the jitter value of each segment of data according to the following formula: ； Among them is the complex frequency domain coefficient corresponding to the k-th point of the spectrum, is the length of the fixed window, represents the jitter value of each data segment, is the natural base, represents the imaginary unit, represents pi, represents the th sampling moment in each data segment; Then calculate the proportion of high-frequency energy in the frequency domain according to the following formula: ； Among them is the set frequency threshold, which determines the starting frequency of the high-frequency part, represents the energy at the k-th frequency point, is the ratio of the high-frequency component energy to the total signal energy, that is, the high-frequency energy proportion.

5. The method according to claim 4, wherein In the said Step 3, the state vector is expressed as: ； Among them, represents the state vector at moment, represents the network delay deviation at moment; represents the queue length at moment; represents the proportion of high-frequency energy in the frequency-domain characteristics at moment. Then regard the state of the network as a linear dynamic system, and represent it through the following state evolution equation and observation equation: ； ； where F is the state transition matrix, denotes the state vector at the previous moment of the moment, B is the control input matrix, denotes the control input at the previous moment of the moment, denotes the process noise at the previous moment of the moment; denote the actual observed value of the moment RTT or the queue length, is the observation matrix, and denote the observation noise.

6. The method according to claim 5, wherein In the state evolution equation, the state transition matrix F is expressed as: ； Among them represents the influence coefficient of queue length on RTT deviation, represents the influence coefficient of RTT deviation on the proportion of high-frequency energy, represents the influence coefficient of RTT deviation on queue length, represents the relationship coefficient between network load and short-term fluctuation, represents the influence coefficient of the proportion of high-frequency energy on RTT deviation, represents the influence coefficient of the proportion of high-frequency energy on queue length; Control input matrix It is expressed as: ； Among them represents the influence coefficient rate of the control input on the RTT deviation, and the influence coefficient of the rate on the RTT deviation represents the adjustment coefficient of the rate on the queue length represents the suppression coefficient of the rate on the high-frequency energy represents the constraint coefficient of the rate on the queue growth represents the coupling coefficient of the rate on the frequency-domain characteristics represents the smoothing coefficient of the rate on the delay fluctuation In the observation equation, the observation matrix is expressed as: ； Among them , and respectively represent the influence coefficients of RTT deviation, queue length, and high-frequency energy proportion on the RTT measurement value; , and respectively represent the influence coefficients of RTT deviation, queue length, and high-frequency energy proportion on the queue length measurement value.

7. The method according to claim 5, characterized in that, In the said Step 3, the real-time network state estimation of the state vector based on the Kalman filter model includes the following steps: Perform real-time network state estimation through two stages: prediction and correction: In the prediction stage, it includes the predicted state and predicted error covariance matrix obtained through the following formula: ； ； wherein represents the predicted value of the state at the moment, and is the optimal state estimate value of the previous moment at the moment; represents the error covariance matrix of the time prediction, represents the error covariance matrix of the prediction at the previous time of the time, superscript of represents the matrix transpose, represents the covariance matrix of the process noise; In the correction stage, it includes first calculating the Kalman gain through the following formula, and then combining the observed data and the predicted state based on the Kalman gain to obtain the optimal state estimation value: ； ； Among them represents the Kalman gain, The superscript of represents matrix transpose, represents the covariance matrix of the observation noise; representing the state estimate with the optimal network state at a moment, is the innovation amount, representing the difference between the observed value and the predicted value; Then update the error covariance matrix through the following formula: ； wherein represents the updated error covariance matrix and is used as the next moment at time k for use represents the identity matrix 8. The method according to claim 7, wherein In the said Step 4, the process of dynamically adjusting the weights of the RTT, queue length, and frequency domain features through the Sigmoid function includes: First, use the Sigmoid function to dynamically map the proportion of high-frequency energy into the weight adjustment coefficient to dynamically adjust the RTT weight, and then further set the queue length weight and the proportion of high-frequency energy weight based on the RTT weight: ； ； ； wherein represents the weight of the RTT, is the sensitivity adjustment coefficient, is the threshold of the high-frequency energy ratio, is the natural base; Indicates the queue length weight; Indicates the weight of the high-frequency energy proportion.

9. The method according to claim 8, wherein In the said Step 4, the adaptive adjustment of the parameters of the PID controller according to the weights includes: The proportional gain of the PID controller is adjusted based on the following equations: , the integral gain , and the derivative gain These three parameters are adjusted respectively as follows: ； ； ； wherein is the initial proportional gain, is the adjustment factor of the proportional gain; is the initial integral gain, is the adjustment factor of the integral gain, represents the control error of the PID controller, is the error threshold, represents other situations; is the initial differential gain, is the adjustment factor of the differential gain.

10. The method according to claim 9, wherein The said Step 5 includes: First, calculate the control error based on the network estimated state : ； wherein, is the currently real-time estimated round-trip delay, and are the target network delay and the queue length at time, respectively, both included in ; is the target queue length; Then, based on calculate the control output of the PID controller : ； wherein represents the mixed error value at the historical moment, represents the control period, represents the mixed error value at the previous moment of the moment; Then, according to adjust the transmission rate : ； ； wherein represents the adjustment value of the transmission rate, is the conversion factor that maps the PID control signal to the actual network control quantity, represents the transmission rate at the next moment of the moment, represents the transmission rate at the moment; Meanwhile, the transmission rate is monitored in real time , when the transmission rate decreases to a preset threshold , an ECN marking operation is triggered to mark the ECN field in the IP header of the data packet, so that the source node actively reduces the transmission rate after receiving the data packet with the ECN mark, thereby completing the Explicit Congestion Notification (ECN).

Images