Ship-borne network performance self-optimization method based on reinforcement learning

Abstract

The invention discloses a ship-borne network performance self-optimization method based on reinforcement learning, and the method comprises the following steps: 1) monitoring the network state of dataflow in real time, collecting the network state, and obtaining the current network state when an elephant flow arrives; 2) identifying the service type of the data traffic; 3) counting the flow entryAction fields to obtain a service flow source / destination address of the elephant flow, and determining a scheduling path set; 4) training a deep reinforcement learning model by taking the network state of the elephant flow, the service type of the flow and the scheduling path set as input; 4) outputting a globally optimal path solution through the calculation of the deep neural network; 5, afterthe global optimal path is determined, generating a new elephant flow forwarding route, and achieving elephant flow rerouting. According to the invention, the global optimal path is calculated for the elephant flow according to the current state of the network and the service flow information, rerouting of the elephant flow is completed, and the utilization rate of network resources can be effectively increased.

Description

technical field [0001] The invention relates to network technology, in particular to a method for self-optimizing shipboard network performance based on reinforcement learning. Background technique [0002] In the ship network, there are various types of services, such as power monitoring system, power monitoring system, damage control monitoring system, ship-wide equipment support management system, driving control system, etc. In the traditional network architecture, the link utilization rate is low, and it is difficult to effectively supervise and control the network, mainly due to the lack of expansion of the core network layer protocol and the tight coupling with the corresponding hardware devices, which leads to the integration of QoS in the network and the access of edge users at any time And a large number of problems exposed in the depth of network management and operation and maintenance. One of the main reasons for congestion, packet loss, etc. in the network is ...

AI Technical Summary

Problems solved by technology

Among them, the host-based detection method checks the traffic on the terminal host, which can obtain higher detection efficiency with lower overhead, but detecting elephant flows on the host is not conducive to the scheduling of elephant flows, thus affecting the overall network. performance

Sampling detection method distinguishes elephant flow from mouse flow by using packet sampling, but needs to send more control messages to notify all relevant devices, increasing system overhead

The aggregated statistical message detection method maps the source and destination IP addresses of each data flow to a two-dimensional space, and then uses the aggregation request method to obtain statistical data until the elephant flow is isolated in a relatively small area, which can reduce Bandwidth consumption, but when the elephant flow is concentrated in a certain area, it needs to be divided into areas, which increases the complexity of detection

The classifier-based detection method runs two classifiers on the switch and the controller to detect the elephant flow, which can improve the detection accuracy of the elephant flow, but at the same time affects the detection efficiency

Images