Estimation of time varying signal parameters using an improved Adaline learning algorithm

This paper presents an improved adaptive linear combiner (Adaline) structure for fast estimation of time varying power signal parameters corrupted by noise. Unlike the conventional Adaline approach, the new algorithm minimizes an objective function based on weighted square of the error and uses a modified recursive Gauss Newton (MRGN) method. The Hessian matrix, obtained by minimizing the objective function, was simplified using certain approximations. A weight adjustment procedure for the Adaline is defined in a decoupled manner for direct current (DC), fundamental, harmonic components and system frequency. The new improved Adaline, thus produces a faster convergence and tracking accuracy for the time varying distorted power system signals. To test the effectiveness of the algorithm, several time varying power network signals were simulated with abrupt change in system frequency, harmonics, decaying dc components with low signal to noise ratio (SNR), and the changing parameters were estimated. The performance of proposed Adaline structure is compared with the standard Adaline structure in terms of accuracy.