Fuzzy neural network is the product of the combination of fuzzy theory and neural network. It combines the advantages of neural network and fuzzy theory, integrating learning, association, recognition and information processing. There is a sharp contradiction between the complexity of the system and the required accuracy. To this end, by simulating human learning and adaptive capabilities, people have proposed the idea of ​​intelligent control. Control theory expert Austrom (1991) pointed out at the IFAC conference that fuzzy logic control, neural network and expert control are three typical intelligent control methods. Usually the expert system is built on the expert experience, not based on the operational data generated by the industrial process, and the inaccuracies and uncertainties of the general complex system are difficult to grasp even by the domain experts, which makes it very difficult to establish an expert system. . Fuzzy logic and neural network are two typical intelligent control methods, each with its own advantages and disadvantages. The fusion of fuzzy logic and neural network--Fuzzy Neural Network absorbs the advantages of fuzzy logic and neural network. Avoiding the shortcomings of both, has become one of the hotspots of today's intelligent control research Fuzzy logic (FL), neural network theory (NN), genetic algorithm (GA), stochastic reasoning (PR), and the combination of confidence network, chaos theory and partial learning theory form a kind of collaboration. This fusion is not chaotic. The fuzzy logic, neural network and genetic algorithm are put together, but the problems in the field are solved by various methods and complement each other, thus forming the cooperation of various methods. In this sense, the various methods are complementary, not competitive. In a collaborative body, various methods play different roles. Through this collaboration, a hybrid intelligent system is created. Both fuzzy logic and neural network are important intelligent control methods. The two kinds of soft computing methods, fuzzy logic and neural network, are combined to complement each other and form a cooperative body--fuzzy neural network. As an important parameter of network congestion control, RTT can reflect the congestion of the network earlier. According to the obtained RTT estimation value, the literature [1] proposes an RTT-driven congestion control algorithm, which has obvious improvement over the congestion control algorithm based on packet loss rate in terms of real-time and network state oscillation suppression. Select formula (1) to estimate the value of RTT: RTTn+1=RTTn+gE (E=RTTm-RTTn) (1) Where RTTm is the currently measured RTT value; RTTn is the average RTT estimate of the previous probe packet, g∈(0,1). Different networks or different time slots of the same network have a great influence on the selection of g. 2] For reliable multicast transmission, a round-trip time estimation strategy based on active network is proposed. With this strategy, reliable multicast protocol can effectively reduce unnecessary control information in the network. According to the network environment, it can be timely and accurately. Determine the packet rate of the incoming network, thereby improving the throughput of the entire multicast group. RTT prediction research is currently a hot issue, and it is meaningful to accurately predict RTT. The literature [3] uses waveform smoothing index and waveform mutation index based on The sliding window weighted average RTT estimation algorithm is used to estimate the RTT value smoothly. The RTT is predicted by the neural network, which achieves good results. However, this is limited to the state where the network is relatively idle. The arithmetic average filtering and BP network are used. Combined approach, predicting RTT, the prediction result is not ideal when the network is congested. This is due to the RTT error value. As the network load increases, it will also increase, because the queue delay and delay jitter will increase significantly as the network congestion increases. In addition, when the network is congested, the packet or ACK packet will be lost, which will lead to the estimation of RTT. The difficulty increases and the estimated RTT value is not accurate, and some fluctuations occur, which leads to the network out of control. Therefore, this paper adopts the combination of low-pass filtering and MBP network. This paper mainly analyzes the characteristics of RTT and finds It has strong high-frequency noise, and adopts RTT prediction strategy combining low-pass filtering and MBP network. Experiments show that even in the case of busy network conditions, good prediction results can be obtained. The performance of the network environment and network equipment has a great impact on data throughput, resulting in strong randomness of data on the network, often manifested as short-term high-frequency noise. Since the communication data flow between two nodes in the network can have many paths, if the path through which each data packet flows is different, the RTT may be different; on the other hand, even if each data packet is via the same path The destination node is reached, but since the network devices in this path are network shared, the data transmission tasks undertaken by the network devices will not be the same when different data packets are passed, which may result in different RTTs. Accuracy can be affected by the randomness caused by this short-term noise. The network conditions and network device performance are relatively stable over a longer period of time. Since network data is generated under the common constraints of both random and stable, filtering is indispensable for studying network data, because at this time we mainly focus on the regularity of network nodes to the network data. . The idea of ​​the low-pass sliding filter algorithm is: take a data between (0, 1), then: The filtering result = (1-a) & TImes; this sampling value + a & TImes; the last filtering result. Its advantage is that it has a good inhibitory effect on periodic interference and is suitable for occasions with high fluctuation frequency. Select a=0.05. The RTT test was conducted on the campus network. The source node and the feedback node are located at Tianjin Vocational and Technical Normal University and Tianjin University of Technology. In the experiment, 200 TCP packets of 10 bytes in size are sent every 100 ms, then the transmission time and the time of receiving the returned result are recorded, and their difference is calculated, and the difference is RTT. T Copper Tube Terminals,Non-Insulated Pin-Shaped Naked Terminal,Copper Cable Lugs Terminals,Insulated Fork Cable Spade Terminal Taixing Longyi Terminals Co.,Ltd. , https://www.longyiterminals.com
RTT Intelligent Prediction Algorithm Based on Fuzzy Neural Network
Introduction to Fuzzy Neural Network