High-voltage Circuit Breaker Intelligent Diagnosis Technology for Mechanical Faults under Power Internet of Things Context

Abstract

Limited by the characteristics of the device itself and the number of fault samples, the research of high-voltage circuit breaker mechanical fault diagnosis technology is relatively difficult. With the advancement of the power Internet of Things construction, the high-voltage circuit breaker fault information of the entire power system can be stored in the cloud effectively, and the representativeness and comprehensiveness of fault samples are solved, so that artificial intelligence technologies that rely on a large number of samples can be applied, then the intelligent diagnosis of mechanical fault of switch equipment can be realized. From the perspective of the construction of power Internet of Things, this paper adopts the most advanced lightweight convolutional neural network and transfer learning to realize the intelligent diagnosis of mechanical fault of high-voltage circuit breaker. Firstly, an efficient lightweight convolutional neural network is constructed based on the principle that the model's depth, width, and resolution are balanced in the convolutional neural network, which avoids the dependence of feature engineering on expert experience. The lightweight convolutional neural network constructed in this paper not only has the highest recognition accuracy on the mechanical data set, but also greatly reduces the calculation and storage overhead of the model and ensures the real-time, fast, and accurate fault signals processing under the power Internet of Things. Then through transfer learning, it can ensure the maximum fault diagnosis accuracy in mechanical fault diagnosis of high voltage circuit breaker under the condition of small fault sample. Experiments prove that the lightweight convolutional neural network and transfer learning intelligent diagnosis method for high-voltage circuit breaker mechanical faults proposed in this paper have a diagnosis accuracy rate of more than 99% on the public rotation data set, and the recognition accuracy rate of migration to the highvoltage circuit breaker with 97.64%. Compared with the traditional method, the accuracy of fault diagnosis is significantly improved, which provides a new feasible idea for the intelligent diagnosis of high-voltage circuit breaker faults under the power Internet of Things.