After reading the materials for a month, I wrote a thesis to read my experiences.

Subject: Application of Intelligent Multi-Agent System in MPLS Traffic Management

1. MPLS Traffic Management
The traffic of a common IP network is the best-effort forwarding mode, and the IP transmission route is not planned in advance. Network bandwidth cannot be reserved in advance.
MPLS-based network traffic management is designed to reserve network bandwidth. Once a physical link fails, it can be quickly switched to another logical link. When a logical link reaches saturation, the reserved bandwidth can be used.

2 Multi-Agent System
A multi-proxy system is a distributed proxy system. Agents are divided into multiple parts and collaborate with each other. Complete automatic management of network nodes.
The selected proxy system is Jade.

3 Ontology
Language in which the ontology is used to define proxy Communication

4. Intelligent proxy
In order to make proxy more intelligent, I have studied various methods to implement intelligent proxy.
At the beginning, we mainly focused on the integration of proxy and Jess expert systems. Jess is a production system based on rete algorithms. Use the agent supporting Jess to implement customized Intelligent Control

Then I also looked at fuzzy control. Jess and fuzzy control are used in combination with jessfu.pdf.

To make agents more self-learning: I have studied neural networks.
Finally, agentacademe is found. The principle of AA is to mine the data recorded by the agent, generate a new decision tree, and convert the decision tree into Jess rules.

5. Expert System
Rule-based reasoning Rule Based Reasoning
The knowledge we are familiar with can be clearly expressed in the if-then format. We can use RBR
Case-Based Reasoning
If we do not know enough about the environment but can clearly restore the environment, we can use CBR.
Neural Network-Based Expert System

In order to better implement accurate reasoning, we consider combining RBR, CBR, and ANN to complement each other.

6. Neural Networks
Neural Networks simulate the neural results of humans and have good self-learning ability. Using Neural Networks can predict traffic.

7. machine learning and Data Mining
There are many algorithms for data mining, including Association, classification and clustering, and of course time series.
Data Mining converts the knowledge to a rule repository and a case library to make the machine smarter.

8 BDIAgent
Another architecture that implements intelligent multi-proxy is PRS. Use Jadex as the BDI model.
In the traditional Jess + Jade method, Jess needs to contain a lot of irrelevant control information. After Jadex is used, Believes, Goals, and Plans are separated to be more clearly defined.

9 practical application
Monitoring Proxy: 1 monitors and records the current traffic
Control Proxy: 1. Determine whether the network is abnormal based on the data recorded by the current monitoring proxy, and make corresponding adjustments based on the results
2. determine whether an exception is about to occur based on past experience, such as congestion, and make corresponding adjustments based on the results
Mining Proxy: 1. Mining Network exceptions based on traffic data
2. Mining proxy behavior rules based on the behavior controlled by the proxy to find the traffic adjustment algorithm with a high success rate.

Trackback: http://tb.blog.csdn.net/TrackBack.aspx? PostId = 567294