RSP-net: an effective framework for automated root cause location in wireless network optimization scene

Currently, mobile communication network has become a very popular way to provide communication service for individual, industry and commerce. In wireless environment, the base stations provide the wireless access function for users, which play a key role in the communication system. With the development of business, the number of base stations also increases a lot, which can achieve several millions in the whole country. Such a large number of base stations is easy to have failure in a single point, leading to the decrease of service quality and customer complaints in relevant area. When the decrease of service quality happens, it is important to find the root cause based on the performance data and alarms of base stations, and optimize the wireless network as soon as possible. Therefore, it is necessary to provide efficient tools to monitor the network performance data and find the root cause. Addressing the problem, in this paper, we propose an effective framework named RSP-net by using deep learning methods for automated root cause location of service quality decline. Based on the unsupervised network performance data and alarms, we first extract the time sequence information for each base station by using pre-training technique, then fix most of neural network parameters for fine-tuning of downstream tasks. By the application of multiple branches for joint learning, finally we can make fluctuation judgement, locate the root cause type and predict the future network performance to prevent further service quality decline. The experiment results based on the network data of real base stations show that our proposed RSP-net achieves precise root cause location, and we get impressive 90.1% accuracy, which is much better than previous methods. Therefore, we can provide maintainers with automated and reliable root cause location in wireless environment.