Fault diagnosis method for high-voltage circuit breaker based on neural network

Abstract

The invention discloses a fault diagnosis method for a high-voltage circuit breaker based on a neural network, and the method comprises the steps: S1, collecting the current, vibration and mechanical travel data and upper and lower opening distance point data of the high-voltage circuit breaker in a normal, shaft-breaking and under-voltage state during switching-on and switching-off; S2, data preprocessing, including vibration data preprocessing and current data processing; S3, taking the vibration characteristics, the current characteristics, the mechanical characteristics and the corresponding fault labels in the opening and closing process of the high-voltage circuit breaker as a data set, randomly selecting 80% of samples in the data set as a training set, taking the remaining samples as a test set, and randomly selecting 20% of samples in the current training set as a verification set; S4, through wavelet neural network and the Wide& Deep model, performing model training, and obtaining an optimal high-voltage circuit breaker fault pre-diagnosis model through the verification set; S5, testing the high-voltage circuit breaker fault pre-diagnosis model by using the test set, and evaluating the accuracy of the fault pre-diagnosis model.

Description

technical field [0001] The invention relates to a neural network-based fault diagnosis method for a high-voltage circuit breaker. Background technique [0002] The circuit breaker can break the circuit in the event of a short circuit or overload in the power system, and is an extremely important device in the power system. [0003] Currently popular circuit breaker fault diagnosis methods include rule-based expert systems and machine learning methods such as traditional BP neural networks, support vector machines, and correlation vector machines. Among them, the vibration signal is an important data for the study of circuit breaker faults. At present, the vibration features are extracted in advance through wavelet packet decomposition, empirical mode decomposition and other signal processing technologies, and then the data set is constructed together with the fault label, and the corresponding model is obtained through supervised learning. Troubleshooting. The data used in...

AI Technical Summary

Problems solved by technology

The data used in this method is relatively simple, and the vibration characteristics are solidified, which cannot represent the rich fault characteristics contained in the vibration signal. The final model is subject to the vibration feature extraction technology and can be widely used.

Images