Vectorization Method for Device Alarms Achieving High General Ability for AI Application to Network Operations

Network and maintenance operations (NWOPs) are being automated. In recent years, there have been attempts to develop artificial intelligence (AI) using machine learning models such as neural networks in order to automate more sophisticated decisions among NWOPs. Equipment alarms (ALMs) are commonly used in many NWOP decisions. To handle ALMs as AI input requires vectorization as other AI inputs, though vectorization employing general one-hot encoding cause generalization performance problem. Its general performance tends to be low despite the large amount of computation. This paper proposes new vectorization method(ALM2Vec) that reflects ALM data and their relational information onto the network topology. ALM2Vec method is able to reduce the amount of calculation, and learning with a high level of generalization performance can be achieved.