Fuzzy control algorithm for congestion information in digital substation communication network

Abstract

Aiming at the problem that the active queue management algorithm can not explicitly control the queue length and the relationship between throughput and delay, a fuzzy information control algorithm based on Active Queue Management for digital substation communication network congestion is proposed. After packet grouping is entered into the router cache queue, the packets that first enter the cache area of the router are preprocessed, and the data packet is processed fairly by using the geometric distribution function. The combination of Smith predictive control and adaptive fuzzy control is used to compensate the network delay of packets, eliminate the negative impact of time delay on active queue jitter and delay jitter, and control congestion according to fuzzy rules intelligently. The experimental results showed that the proposed algorithm can maintain smaller queue oscillations, especially when the network conditions change. It can effectively eliminate the impact of time delay on queue jitter and delay jitter, and improve the overall performance of the network.

Keywords

Digital substation communication network congestion information fuzzy control algorithm

1 Introduction

The smart grid is the latest trend in the development and transformation of the energy industry in the world, which represents the direction of the future development of the power grid. As one of the important components of the smart grid, the importance of communication is becoming more and more important, and more and more problems are produced [1]. Network congestion is one of the most important problems with the development of communication network in digital substation. Since the problem arises, some experts and scholars have not interrupted the research. It has been a hot topic in the field of smart grid research [2]. There are many reasons for network congestion: for example, a large number of storage space in the network is occupied, the bandwidth available in the link cannot meet the current needs, the factors that have not been considered when designing the network congestion control mechanism, and the intentional destruction of some network users. When the network is congested, the performance of the whole link will be worse than before, and packets sent by the sending end can’t be in time or can’t reach the receiver. When the sender receives a ACK response to the receiving end, it will exceed the timeout retransmission time. c When congestion is serious, it can also paralyze the entire network and is unable to use. In the current digital substation communication network, we cannot avoid the occurrence of congestion. It is needed to control by a certain way to minimize the possibility of congestion. In this case, the research on congestion control has important theoretical significance and application value.

There are a lot of domestic scholars have achieved many achievements in network congestion information control algorithms, such as Shi et al proposed the network congestion control algorithm based on caching interaction. Through the use of routers in the network cache, it dynamically expands the buffer size and controls the transmission rate of the data packet logically. At the same time, combined with the existing network congestion control algorithm, the packet transmission rate threshold is dynamically adjusted to smooth the burst traffic and alleviate the network congestion [3]. The queue jitter of this algorithm is very large. After generating or reducing burst traffic, it will take a long time to reach steady state and stable time. Wen et al. Proposed a congestion control algorithm based on improved congestion traffic allocation. After the measure of energy consumption and transmission delay of all paths, the path of a number of low energy consumption is chosen, to detect congestion in the network area through the method of setting threshold and forecast, once the congestion occurs, it uses the way of rational redistribution of traffic to enable nodes to recover faster from congestion and ensure that the data of congestion area can be transferred to non congested area as quickly as possible [4]. The algorithm has less queue jitter and effectively improves the link utilization, but increases the data packet loss rate at the same time. Qiu et al proposed a congestion control algorithm with the ability to predict and self adjust. It is used to predict network congestion by fuzzy neural network controller and real-time prediction based on queue length in buffer. Before the congestion occurs, the sending rate is controlled by restraining the transmission rate of the input terminal, and adjusting the transmission rate dynamically by combining the incremental and decreasing parameters [5]. This algorithm is more effective for deadlock and global synchronization, but the link utilization rate is low and the delay is increased. Huang et al proposed a congestion control algorithm based on substream packet loss differentiation and sharing bottleneck. It dynamically adjusts congestion window by detecting the bottleneck of network neutron traffic sharing, and makes full use of network resources at non shared bottleneck links. When multipath transmission control protocol is applied to heterogeneous networks, packet loss differentiation mechanism is introduced to avoid wasting network resources due to non congestion packet loss [6]. This algorithm effectively reduces the packet loss rate, but performs poorly on the performance of queue jitter and average delay time.

2 Material and methods

2.1 Active queue management algorithm based on pretreatment

2.1.1 Congestion detection method

The delay jitter in the digital substation communication network generally refers to the difference between the time delay of the current time and the average delay of the network link. The congestion in the network can be judged by the change of the delay jitter and the queue length of packet grouping. It is assumed that the bandwidth of the link in the communication network is fixed value $\dot{C}$ . At a certain time t, the actual leader in the network is L_t, and the transmission rate is v_t. Then the value of the delay d in the network link is: $d = L_{t} / v_{t}$ (1)

The data traffic in the communication network of digital substation has mutagenicity. To calculate the time delay jitter, we must first calculate the average delay value. The average delay avg _ d is calculated by weighting factor Wd. $avg_d = j_d \times (1 - Wd) + d \times Wd$ (2)

The value of the delay jitter j _ d is as follows: $j_d = d - avg_d$ (3)

In the network transmission, the time delay fluctuation determines the size of the delay jitter. The time delay jitter determines the quality of the network QoS. In order to ensure the network QoS, the queue length and the sending rate are balanced.

The real-time queue length in the buffer time assuming m router is L_m, L max is the maximum length of the buffer to store data packets, by using Formula (1) to (3) to calculate the delay and jitter of delay jitter, if the calculated delay jitter is zero, It shows that the size of the queue length at every moment of packet grouping arrival in the cache area is fixed. So the transmission rate is v_t, which is also fixed value and the transmission rate must be less than the transmission rate of the link. If the delay jitter is not zero, it indicates that the actual queue length reaches the buffer in the unceasing change, in order to control the sending rate to burst in a short period of time and control the network congestion, this algorithm will recalculate the queue length. If the recalculated queue length is smaller than L max/2, it indicates that the arrival rate is larger than the transmission rate. Digital substation communication network is a dynamic real-time transmission mode, burst at any time arrival will cause the router queue length buffer packet increases. At this point, it needs to adjust the transmission rate of packets as follows: $L_{m} / v_{m} - avg_d = 0$ (4)

The transmission rate v_m of the m time is generally adjusted by the delay jitter obtained from the difference between the current time delay and the difference of the preceding time delay. $L_{m} / v_{m} - L_{m - 1} / v_{m - 1} = 0$ (5)

Where, L_m-1 represents the real-time queue length in the m - 1 moment of the router cache, and v_m-1 represents the transmission rate of the m - 1 time. The transmission rate of the packet can be increased to: $Δ v = (L_{m} / L_{m - 1} - 1) \times v_{m - 1}$ (6)

The maximum bandwidth of a network link is C″, and the maximum transmission rate of packet cannot exceed the bandwidth C″ of the network link. If the delay jitter of the network link is greater than zero, the queue length and the calculated is greater than L max/2, it can be determined as the network has occurred congestion, then the use of preprocessing algorithm [7], geometric distribution function hit algorithm [8] and the complex curve function algorithm [9] obtained the corresponding packet loss probability; when the delay jitter of the network link the queue length is less than zero, while the calculated is less than L max/2, then to reduce the sending rate of source packets: $Δ v = (1 - L_{m} / L_{m - 1}) \times v_{m - 1}$ (7)

2.1.2 The judgment and treatment of fairness

Assuming that the network link bandwidth is C″, there are N sending ports at the source end of the link to send TCP streams, and one sending ports to send UDP streams. The destination ends have N receivers receiving TCP streams, and one receiver to accept UDP streams. b_i is the packet size of router buffer i, p_i is the total packet loss probability, b is the total size of all enter the router buffer packets (hypothesis there are N packets into the router cache), h_i is the matching probability (h_i = b_i/b), τ is the delay TCP traffic, x_i is the source transmission rate of traffic II, r is the packet loss probability for composite curve function algorithm. Because the hit probability accords with the Bernoulli probability model [10], assuming k packet was hit, then the hit probability is $h_{i}^{k} (1 - h_{i})$ , the (k + 1) packet was not hit, then the network link in total packet loss probability is: $p_{i} = r + \sum_{k = 1}^{\infty} (1 - r) (k + 1) h_{i}^{k} (1 - h_{i}) = r + (1 - r) \frac{2 h_{i} - h_{i}^{2}}{1 - h_{i}}$ (8)

The total packet loss probability is obtained from the direction that cannot be hit by a geometric distribution function. Assume that there are i arbitrary packets in the network link into the router cache, then any probability of a packet being hit is (1 - r) (1 - h_i), after the n hypothesis, the first n - 1 have been hit, the n was not hit, then the probability of the occurrence is $h_{i}^{n - 1} (1 - h_{i})$ , it conforms to the geometric distribution of q_i = 1 - h_i, the packet that reaches the cache is not hit 1/q_i = 1/(1 - h_i) times. It is assumed that the delay is τ, and there are x_i (1 - r) τ packets entering the router buffer in time τ. During that time, τx_i (1 - r)/(1 - h_i) packets are not hit. Also, because the probability that the packets arriving at the router cache are not hit is (1 -1/b), the probability that the packet is not discarded is: $\begin{matrix} 1 - p_{i} = \\ (1 - r) (1 - h_{i}) (1 - 1 / b)^{τ x_{i} (1 - r) / (1 - h_{i})} \end{matrix}$ (9)

Assuming that the network is in full link state at the time, x₀ represents the source end transmission rate of the UDP traffic at this time. x₁ indicates the source port transmission rate of the first TCP stream, and the transmission rate of each TCP stream is equal to that of the source port. Then: $x_{0} (1 - p_{0}) + {Nx}_{1} (1 - p_{1}) = C$ (10)

From the above, the number of TCP adaptive traffics in the router cache is Nb₁ = b - b₀ = b (1 - h₀). Assuming that the packet loss rate of the compound curve function is r, $(h_{0}^{2} - 3 h_{0} + 1) / (1 - h_{0})$ represents the probability that the corresponding geometric distribution function is not discarded in the hit algorithm, and the UDP traffic can be obtained from the upper form. When the hitting probability is h₀, the packet loss probability of the set distribution function is: $p_{p} = 1 - \frac{h_{0}^{2} - 3 h_{0} + 1}{1 - h_{0}} = \frac{2 h_{0} - h_{0}^{2}}{1 - h_{0}} max p$ (11)

Among them, max p represents the critical value of the network from mild congestion to severe congestion.

On the geometric distribution function for the hit algorithm of packet loss probability, the size of the TCP adaptive stream p_p that enters the router cache is basically zero, to protect the adaptability of TCP traffic, and the fairness between various types of streams in modern communication networks is well enhanced [11].

2.2 Adaptive fuzzy network congestion control algorithm based on fuzzy Smith system

Network congestion is detected mainly in the digital substation communication network by detecting the average queue length of packet packets, it combines the preprocessing algorithm and the geometric distribution function hit algorithm to get the packet loss probability of packets, and carries out the implementation of the corresponding packet loss strategy. The algorithm mentioned above has greatly improved in avoiding global synchronization, preventing deadlock, and continuous full queuing [12], but because max p is a fixed value, if the value is too large will make use of network congestion in mild severe packet loss strategy when the network congestion, continued in the light load state, resulting in a tremendous waste of cyber source; if the value is too small to make congestion control is too conservative, the dropping probability is smaller, the router cache queue is overtraffic, resulting in network performance is greatly reduced. In view of the network delay data packet, it uses an adaptive fuzzy control algorithm for active queue management algorithm based on fuzzy logic, combines Smith predictive control with adaptive fuzzy control, eliminates the negative impact on active queue jitter and delay jitter of the ring, and intelligent controls the congestion based on fuzzy rules (as shown in Fig. 1) [13 –17].

Fig. 1

Fuzzy Smith system structure.

Among them, q_b represents the queue length, p (t) represents a dynamic marking probability, q (t) represents the target queue length, F (s) represents fuzzy controller, e^−s represents the pure time delay, G_w (s) represents the transfer function of the sending window, G_q (s) represents the transfer function of buffer queue length, (1 - e^−s) G_w (s) G_q (s) represents the Smith compensation function. In the actual substation communication network, the instantaneous queue length in the router can be obtained in real time. When Q_r is chosen as the expected queue length, the instantaneous queue length deviation e (*) and the deviation rate Δe (t) can be expressed as follows: $[\begin{matrix} e (t) = Q_{r} - Q (n) \\ Δ e (t) = de (t) / dt \end{matrix}$ (12)

e (t) and Δe (t) are selected as the input of the fuzzy controller. These two inputs can strictly reflect the dynamic characteristics of the queue length change in the router. In fuzzy control system, E and Ec are used to represent their corresponding linguistic variables, respectively. Output is chosen as Δp (t) of dropping probability, and its linguistic variable in fuzzy system is Dp. In the fuzzy set, the more the elements in the set, the more precise control, but the amount of calculation will increase exponentially. In order to better respond to the speed of processing, based on the expert knowledge and experience of the designer, balance control precision and computation, the set of fuzzy sets for E and Ec here were [NB (negative big), NS (negative small), ZE (zero), PS (positive small), PB (positive big)], for the convenience of calculation, the membership degrees of the fuzzy sets are all triangular functions, Q_E is queue length of the quantized domain. The control rules of the fuzzy controller are given by using Table 1.

Table 1

Control rules for fuzzy controllers

E
D_p		NB	NS	ZF	PS	PB
	NB	3	3	2	1	0
	NS	2	2	1	0	−1
EC	ZF	1	2	0	−1	−1
	PS	0	1	−1	−2	−2
	PB	−1	0	−2	−3	−3

According to the fuzzy rules above, the relationship between the input and output of the fuzzy system can be expressed as follows:. $\begin{matrix} Δ p (n) = \\ \sum_{η = 1}^{m} (y^{- η} u_{l} (e) Δ p (t)) / \sum_{i = 1}^{m} (u_{l} (e) Δ p (t)) \end{matrix}$ (13)

Among them, y^−η represents the central value of the fuzzy set of the “then” part of the η fuzzy rule, and u_l (e) represents the membership function of the fuzzy set η.

For two-dimensional fuzzy controller, its control parameters include quantization factor k_e, k_c and the scale factor k_u. When the network parameters change, the gain of fuzzy logic controller can be adjusted to counteract the influence of the change. When the number $\dot{N}$ of network connections increases, the p (t) should be increased. When the link capacity C increases, p (t) should decrease, while the increase of delay will increase p (t). Accordingly, the control behavior of fuzzy logic controller is increased or reduced, and the performance of fuzzy system is improved. In order to correct the control parameters and modify the control rules, we should evaluate the control performance and set the fuzzy performance index FP for the control quality of the evaluation system as: $FP = min ∖ {u_{OV} (e_{OV}), u_{RT} (e_{RT}) ∖}$ (14)

Among them, OV and RT are the overshoot and rising time of the system response, respectively. The deviations of OV* and RT* are RT* and e_RT, respectively. When the number of network connections increases, the overshoot will increase. At the same time, the control behavior should be increased. Otherwise, the control behavior should be reduced. When the link capacity increases, the overshoot will decrease, so the control behavior should be reduced. The correction rules for specific parameters are as shown in Table 2. The changing intervals of these rules are [- c_i, c_i] and [- d, d], in which Δc_i and Δd denote the increase or decrease of c_i and d,respectively.

Table 2

Regulation rules of ratio coefficient

Output characteristic	Input characteristics	Δc₁	Δc₂	d
e_OV	N	NB	PS	PB
	P	PB	NS	NB
e_RT	N	PB	NB	NB
	P	NB	PB	PB

The Δc_i and Δd can be calculated by the above rules and the simplified fuzzy reasoning, so the parameters of the fuzzy controller can be adjusted according to the formula as follows: ${\begin{matrix} {\hat{c}}_{i} = c_{i} + (1 - FP) Δ c_{i} \\ \hat{d} = d + (1 = FP) Δ d \end{matrix}$ (15)

Among them, ${\hat{c}}_{i}$ and $\hat{d}$ represent the updated values of c_i and d respectively. Through the above adjustment process, the fuzzy controller can dynamically respond to the change of network parameters. Set the output for a given network queue the expected response as $q_{k}^{*}$ , with the expected response to evaluate the sampling quality degree of control response in each moment, when the control response obtained at each sampling point, the system response is consistent with the expected response by modifying the control rules that are related to the present state of control by real time learning algorithm. The modified control rule is to modify the real number u_i of the result item of the control rule at each sampling point, and the new value is obtained by the next calculation. ${\hat{u}}_{η} = u_{η} + (1 - FP) Δ u \times w_{η}^{k}$ (16)

Among them, $w_{η}^{k}$ represents the fitness of rule η at k sampling time, ${\hat{u}}_{η}$ represents the correction value of u_η, and the control increment of the system at k time can be calculated by the next type. $Δ p_{k} = \sum_{η = 1}^{5} w_{η}^{k} {\hat{u}}_{η} / \sum_{η = 1}^{5} w_{η}^{k}$ (17)

He control amount can be calculated from the control increment of the k time: $p_{k} = p_{k - 1} + Δ p_{k}$ (18)

3 Results

In the evaluation of congestion control algorithm, the following performance indicators were compared to evaluate the performance of the algorithm: the average queue length, queue jitter, throughput, packet loss probability and queuing delay. The queue length is basically proportional to the queuing delay. In general, network latency is composed of two parts: transmission delay DTime_i′ and queuing delay ETime_i′. The “dumbbell” topology is used in the simulation, the transmission delay is generally unchanged, and the queuing delay on the bottleneck link is mainly observed. The transmission delay of the packet in the network is constantly changing, and the queue length jitter affects the queuing delay and the transmission delay of the whole packet. For some substation communication real-time services, the effect of queue jitter on QoS is more obvious. The queue jitter should be reduced as far as possible, and formula for calculating queue jitter by standard variance is as follows: $D = \frac{\sum_{i^{'}}^{n^{'}} ({DTime}_{i^{'}} - {ETime}_{i^{'}})}{n^{'}}$ (19)

Among them, n′ is the total number of packets that are transmitted within the sampling time. $DD = \sqrt{\sum_{i^{'}}^{n^{'}} (x_{i^{'}} - \bar{x})^{2} / (n^{'} - 1)}$ (20)

Among them, DD represents the queue jitter, $\bar{x}$ represents the captain average, x_i′ represents the total sampling number.

Effective throughput is another important indicator of the performance of the evaluation queue. Effective throughput refers to the total byte of all the actual data transmitted through the queue in the unit time (sampling frequency), rather than the total byte of all the packets. As the following formula: $GP = \frac{df}{t^{'} \times 10^{6}}$ (21)

Where, df represents the total byte of the unit time transmission, and t′ represents the sampling time.

Packet loss rate will reflect the actual loss of algorithm packets, and the utilization rate of links is directly related to the efficiency of network transmission. They are all important evaluation indicators. As the following formula: $DB = \frac{\sum g}{\sum h}$ (22) $LL = \frac{\sum f^{'}}{\sum e^{'}}$ (23)

Where, the DB represents the packet loss rate, g represents the bytes discarded by the queue, h represents the byte of the queue. LL represents link utilization, f′ represents the bytes sent by the bottleneck link, and e′ represents the maximum bytes that the bottleneck link allows to send.

In order to balance throughput and delay, achieve high throughput transmission and maintain smaller queuing delay, a custom evaluation parameter c′ is set. The parameter is the ratio of effective throughput to queuing delay in unit sampling time. The larger the c′ value, the better the algorithm. It means that the throughput is larger under the same throughput or with the same delay. $c^{'} = GP / D$ (24)

The method of taking the average value is adopted for the overall evaluation of the algorithm, and the formula is as follows: $c^{*} = \sum_{i^{'}}^{n^{'}} c^{'} / n^{'} = \sum_{i^{'}}^{n^{'}} \frac{GP / D}{n^{'}}$ (25)

3.1 Burst traffic

The network congestion control algorithm for interactive cache queue based on jitter is very large from the experimental results. After increasing or reducing burst traffic, it will reach steady state only after 80 s, and the stability time is relatively short, 130 s to 175 s achieves a relatively stable state. The instantaneous performance of the algorithm is poor. It dynamically adjusts the congestion window by detecting the sharing bottleneck of the network neutron traffic. The average queue length and jitter of the corresponding queue are relatively large, the average queue length is 500pkts, and the jitter is 380pkts. The average queue length of the network congestion fuzzy control algorithm based on active queue management is 410, which is very close to the expected queue length. The jitter of the queue is only 50, indicating that the algorithm is stable and robust. The minimum queue jitter based on improved congestion traffic allocation congestion control algorithm is 100.85. The queue length of the algorithm under burst traffic is changed as shown in Fig. 2.

Fig. 2

Queue length of different algorithms under burst traffic.

The average packet loss probability of the network congestion control algorithm based on cache interaction, the network congestion control algorithm based on improved congestion traffic allocation and the network congestion control algorithm with self adjustment ability is lower. The packet loss probability of the network congestion fuzzy control algorithm based on active queue management is the least, and the basic stability is about 2%. The packet loss probability of the burst traffic algorithm is shown in Fig. 3.

Fig. 3

The packet loss rate of different algorithms under burst traffic.

The sudden traffic experimental results were collected, which summarizes the average queue length (pkts), queue jitter (pkts), effective throughput (Mbps), link utilization rate (%)and custom evaluation parameters (Mbps/s) as the evaluation indexes are shown in Table 3.

Table 3

Analysis of the results of the burst traffic experiment

Algorithm Type	Average Captain/(pkts)	GP/(Mbps)	Link utilization /(%)	Delay /(ms)	Packet Loss /(%)	Evaluation Parameters/(Mbps/s)
R	600	14.25	94.58	179.52	4	79.54
P	500	14.25	95.32	175.25	4	65.25
O	450	14.25	96.58	214.34	3	68.58
I	400	14.45	97.52	120.25	2	89.63
K	450	14.41	96.25	165.25	5	84.21

Table 3 showed that the average queue length of the network congestion fuzzy control algorithm based on active queue management is the closest to the expected queue length, and the error is minimal; compared with other algorithms, the queue jitter and average delay time is minimum and the evaluation parameters of the highest; it is proved that fuzzy congestion control algorithm based on active queue management with fuzzy control mechanism can effectively control queue length and jitter in the burst traffic experiment, balance the relationship between throughput and delay, and improve the performance of the algorithm as a whole. Because the concept of fuzzy control mechanism is introduced, the network congestion fuzzy control algorithm based on active queue management is superior to other network congestion control algorithms in queue shaking, link utilization, average delay, average packet loss probability and overall evaluationparameters.

3.2 Non response traffic

Because of no congestion control mechanism, the non response traffic cannot reflect the network congestion, which makes some algorithms fail. Although the current UDP traffic in the network only accounts for 10% –20%, the non response traffic is an important research topic to ensure network stability. In the experiment, 50 sets of FTP connections are also set as background traffic. In the 80 s, 50 groups of UDP connections are opened to send CBR traffic, and all UDP connections are closed at 240 s. UDP connections are also divided into 5 categories according to the delay. The results of the non response traffic experiment are shown in Table 4.

Table 4
Analysis of experimental results of non response traffic

Algorithm Type Average Captain /(pkts) GP/(Mbps) Link utilization /(%) Delay /(ms) Packet Loss /(%) Evaluation Parameters /(Mbps/s)

R 528 14.25 91.25 182.51 5 80.69

P 519 14.29 95.51 193.45 5 79.64

O 432 14.28 96.54 205.34 5 65.36

I 400 14.45 97.74 122.58 3 91.36

K 499 14.22 96.41 188.78 4 88.64

Algorithm Type	Average Captain /(pkts)	GP/(Mbps)	Link utilization /(%)	Delay /(ms)	Packet Loss /(%)	Evaluation Parameters /(Mbps/s)
R	528	14.25	91.25	182.51	5	80.69
P	519	14.29	95.51	193.45	5	79.64
O	432	14.28	96.54	205.34	5	65.36
I	400	14.45	97.74	122.58	3	91.36
K	499	14.22	96.41	188.78	4	88.64

The results of different congestion algorithms in the non response traffic experiments are summarized in Table 4, and the following conclusions are obtained.

The average queue length and the expected length error of the network congestion fuzzy controlalgorithm based on active queue management is small; compared with other algorithms, minimum average delay, it has the lowest queue jitter and average delay, and the highest evaluation parameter, effective throughput and link utilization rate. It is proved that the proposed algorithm is stable,timeliness and effective for non response flow control, and ensures high throughput and link utilization.

3.3 Short life traffic

At present, there are also a large number of short lifetime traffic in the digital substation communication network, such as HTTP and TELNET applications. These traffics have shorter survival time and less transmission data, but if a large number of short life time traffic are generated at the same time, they will have some impact on the network. The experiment uses Short-lived FTP as test traffic, and these FTP traffic are also divided into 5 categories, generating 20 FTP connections per second, sending 3 packets per connection, adding Short-lived FTP to 80 s. The test time is 15 s, and the test traffic amounts is 150 groups. The results of the short life traffic experiment are shown in Table 5.

Table 5
Analysis of the results of the short life traffic experiment

Algorithm Type	Average Captain /(pkts)	GP/(Mbps)	Link utilization /(%)	Delay /(ms)	Packet Loss /(%)	Evaluation Parameters /(Mbps/s)
R	522	15.25	89.69	199.65	4.78	88.36
P	517	15.27	91.69	195.67	4.65	79.69
O	469	15.36	91.25	192.69	4.35	75.48
I	411	14.45	96.57	121.36	2.69	85.69
K	436	15.26	93.65	188.36	3.89	90.36

From Table 5, the results of different algorithms in short life traffic experiments are summarized, and the following conclusions can be obtained.

The average queue length and expected length error of congestion control algorithm based on cache interaction is less than that of congestion control algorithm based on sub-traffic packet loss and shared bottleneck.; compared with the other algorithms it has the minimum queue jitter and average delay and the maximum packet loss rate. In terms of the overall evaluation parameters, the effective traffic throughput, link utilization and evaluation parameters of the network congestion fuzzy control algorithm based on active queue management are the largest. It is proved that the fuzzy congestion control algorithm based on active queue management has improved the queue length, queue jitter and overall performance under short lifetime traffic compared with other network congestion control algorithms.

4 Discussion

The performance of the proposed network congestion control algorithm based on active queue management is verified by experiments. Three experimental results clearly showed that the algorithm can effectively control and stabilize the queue length in the expected queue, and the jitter of the queue is relatively small; by introducing the fuzzy control mechanism, it ensures the throughput and delay time to reach equilibrium, and optimizes the shortcomings of the network congestion control algorithm of the original queue management, it is proved that the improved algorithm has the advantages of high efficiency, robustness and stability.

Three experimental results showed that compared with other relative algorithm of network congestion control algorithms, the proposed algorithm has been improved the performance parameters and improved the defects of timeliness and efficiency of the performance of network congestion control algorithm based on active queue management, which has good timeliness and efficiency.

5 Conclusions

With the further research and application of active queue management algorithm, most of the algorithms are rely on intuitive heuristic design, which is blind and one-sided. Most algorithms do not analyze the algorithm itself in a comprehensive and systematic way. Aiming at the problem that the active queue management algorithm can not explicitly control the queue length and the relationship between throughput and delay, a fuzzy control adaptive virtual queue algorithm is proposed. The algorithm is implemented in network simulator NS2, and the stability and robustness of the network congestion fuzzy control algorithm based on active queue management is verified through experiments compared with other algorithms. Aiming at the problem that the active queue management algorithm cannot update the reasoning rules dynamically, the fuzzy control mechanism is introduced into the active queue management algorithm, and an active queue management is proposed. The experiment proves that the network congestion fuzzy control algorithm based on active queue management has a timeliness and effectiveness in updating the rules.

References

Duan

, Shah-Mansouri

, Wang

et al., D-ACB: Adaptive congestion control algorithm for bursty M2M traffic in LTE networks, IEEE Transactions on Vehicular Technology65(12) (2016), 9847–9861.

Yan

H.C.

, Zhang

Q.J.

and Sun

, On congestion control strategy for space delay/disruption tolerant networks, Journal on Communications37(1) (2016), 142–150.

Shi

S.S.

, Ren

Y.M.

, Li

et al., A content store-based congestion control algorithm for named data networking, Chinese High Technology Letters26 (2016), 359–366.

Wen

H.Y.

and Huo

W.D.

, Wireless sensor network traffic reallocation congestion control algorithm, Journal of University of Electronic Science and Technology of China46(2) (2017), 407–411.

Qiu

Z.M.

and Wan

Z.P.

, Congestion control algorithm with prediction and self-regulation ability under complex network, Computer Engineering43 (2017), 115–119.

Huang

H.C.

, Lu

W.J.

and Liu

J.X.

, M PTCP congestion control algorithm based on packet loss differentiation and shared bottleneck link, Computer Engineering and Design37 (2016), 571–576.

Kamisiński

, Domżal

, Wójcik

, et al., Two rerouting-based congestion control algorithms for centrally managed traffic-oriented networks, IEEE Communications Letters20(10) (2016), 1963–1966.

Xiao

, Mao

and Tugnait

J.K.

, MAQ: A multiple model predictive congestion control scheme for cognitive radio networks, IEEE Transactions on Wireless Communications16(4) (2017), 2614–2626.

Xiao

, Zheng

W.X.

, Jiang

et al., Stability and bifurcation of delayed fractional-order dual congestion control algorithms, IEEE Transactions on Automatic ControlPP (2017), 1–1.

10.

Peng

, Li

, Zhang

, Ge

, Mo

, Li

and Ohkoshi

, Characteristics of antibacterial molecular activities in poplar wood extractives, Saudi Journal of Biological Sciences24(2) (2017), 399–404.

11.

Wang

, Zhang

D.Q.

, Su

, Dong

J.W.

and Tan

S.K.

, Assessing hydrological effects and performance of low impact development practices based on future scenarios modeling, Journal of Cleaner Production179 (2018), 12–23.

12.

Qadir

M.I.

, Mushtaq

and Mobeen

, Study of potential carboxylic acid derivatives as D-glutamate ligase inhibitors in, Kuwait Journal of Science45(1) (2018), 100–107.

13.

Favaretto

, Grosset

and Viscolani

, Advertising a social event in a heterogeneous market, Journal of Interdisciplinary Mathematics19(5–6) (2016), 999–1014.

14.

Al-Hassan

Q.M.

, On some interesting properties of a special type of tridiagonal matrices, Journal of Discrete Mathematical Sciences and Cryptography20(2) (2017), 493–502.

15.

Imran

, Raffoul

, Usman

and Zhang

, A study of bifurcation parameters in travelling wave solutions of a damped forced korteweg de vries-kuramoto sivashinsky type equation, Discrete and Continuous Dynamical Systems-Series S11(4SI) (2018), 691–705.

16.

Plata

S.A.

and Sáez

S.B.

, After notes on Chebyshev’s iterative method, Applied Mathematics and Nonlinear Sciences2 (2017), 1–12.

17.

Zhou

, Xu

and Xu

, A sufficient condition for the existence of a k-factor excluding a given r-factor, Applied Mathematics and Nonlinear Sciences2 (2017), 13–20.

Fuzzy control algorithm for congestion information in digital substation communication network

Abstract

Keywords

1 Introduction

2 Material and methods

2.1 Active queue management algorithm based on pretreatment

2.1.1 Congestion detection method

Table 5 Analysis of the results of the short life traffic experiment

5 Conclusions

References

Table 5
Analysis of the results of the short life traffic experiment