The Research Of Multi-State Network Equipment Fault Prediction Based On Neural Network

With the continuous expansion of the network, the network devices running on the network, such as routers, switches and other equipment increased, to ensure network uptime, maintenance of network equipment is not faulty, after a fault to quickly and accurately locate and troubleshoot problems, for network maintenance and management staff is a big challenge.In order to overcome the shortcomings of traditional repair methods, with advances in condition monitoring and fault diagnosis technology, gradually developed a new method of repair-Condition Based Maintenance (CBM). The maintenance mode integrated use of various techniques to obtain information about the device’s operating status and the use of data analysis and decision-making techniques to repair equipment status in real-time or periodic evaluation, the final decision-making scientific maintenance, achieved through state monitoring and forecasting impending fault, establish a reasonable maintenance decisions. Fault prediction technology is an important part of the fault diagnosis technology, is of historical value and the current fault characteristics were analyzed to predict the future value of the fault characteristics to predict the equipment operating status and future trends over time to determine equipment the alert level for possible equipment fault prediction, provide the basis for early prevention and repair faults, have important theoretical value and practical significance of the project.This paper presents a fault prediction method based on neural networks, the introduction of state-based maintenance technology to build a network operating equipment fault prediction model based on multi-state. The model will be based on the severity of the fault warning level is divided into four layers, for different warning levels, neural networks were constructed to solve the prediction accuracy is not high equipment fault problems, improve the ability to predict faults based on multi-state. By collecting information on network equipment operating characteristics, features of the device information obtained sample set, the application design is complete neural network training sample set, and further optimize the structure of neural network design, build fault prediction model based on neural network to realize the equipment fault predictive analysis.Based on the state of maintenance of equipment status based primarily on information obtained predict equipment (or parts) of remaining life, to a certain optimization criteria for the target device to make a maintenance decision, that is, whether the need for preventive maintenance of equipment, if necessary, when It is most suitable for repair. This way of repair maintenance interval is not fixed, its biggest feature is based on the specific status of each device, equipment fault before repair. For devices, based on the state of repair can reduce maintenance and support costs, improve equipment availability and mission success rate; By reducing spare parts, support equipment, maintenance and other support human resource needs, reduce maintenance and support costs; through reduced maintenance, particularly unscheduled maintenance times, reduce maintenance time and improve operational readiness; through condition monitoring, reducing the risk of mission fault caused the process to improve the success rate of the task, providing a great service efficiency.