Parameter identification of key components of thyristor converter valve based on wavelet packet key feature extraction and Elman neural network

Abstract

Online health monitoring of thyristor converter valve is very important for the stable operation of high voltage direct current (HVDC) system. However, in practical projects, only offline annual maintenance can be realized. This paper proposes a method for parameter identification of thyristor valve based on wavelet packet key feature extraction and Elman neural network. By preprocessing the measurable characteristic parameters of thyristor valve in practical engineering, such as wavelet packet time-frequency domain feature extraction, variation coefficient optimization and so on, the characteristic vector that can characterize the health state of thyristor valve is obtained and sent to Elman neural network for training, which realizes the rapid online identification of multiple electrical parameters of thyristor modules in converter valve. Simulation and comparative experiments show that the key feature extraction method proposed in this paper has significant advantages in identification accuracy, and can better identify the damping capacitance and leakage resistance of all thyristor modules. Among them, the deterioration of the damping capacitance is more easily identified, the identification accuracy is the highest and the relative identification error is less than 5%. The parameter identification method proposed in this paper has the advantages of fast identification speed, strong model generalization ability and wide application. This provides a new idea for the online condition monitoring of thyristor valves.