Fault Early Warning and Location of Dry-Type Air-Core Reactor Based on Neural Networks

In the power system, dry-type air-core reactor can protect the power system from major damage when a short circuit happens. In general, the earlier the fault is detected, the less damage it will cause. Most of the existing methods only focus on the whole temperature change of the equipment, and ignore the local changes, which may hide the early signs of fault and make it is difficult to provide immediately early warning. In addition, when fault happens, the location information of the fault is extremely important for maintainers to maintain equipment. However, most of the current methods can only give the running status of the equipment, but not the fault location. In this paper, we propose a new method to diagnose dry-type air-core reactor faults. This method is a hierarchical multi-level neural network that includes sub-networks and main networks, where sub-networks are used to infer the status of each region, and the main network is used to infer the status of whole equipment. Our method not only can give warn of faults early, but also can provide the specific location of the fault in time. The verification experiments show that proposed method is feasible and effective.