Parameter Estimation of Decaying DC Component via Improved Levenberg-Marquardt Algorithm

Fault current usually contain a decaying DC component and some kinds of noise. This DC component and noise decrease the accuracy and speed of the operation of digital relay protection. In order to remove the decaying DC component and noise in current signals for power system, parameters of decaying DC component should be estimated firstly. To solve this parameter estimation problem, a specific neural network is proposed, and then an adaptive learning algorithm based on improved Levenberg-Marquardt algorithm is derived to iteratively resolve its weights by optimizing the pre-defined objective function. From weights of the trained neural network, all parameters of decaying DC components can be well calculated. Profiting from good nature in fault tolerance of neural network, the proposed algorithm possess a good performance in resistance to noise. Simulation experimental results indicate that our algorithm can achieve a high accuracy with acceptable time consumption for parameters estimating in noise.