Broad Learning System Using Rectified Adaptive Moment Estimation for Harmonic Detection and Analysis

The extensive application of nonlinear components and power electronic devices in power systems has led to increasingly deteriorating harmonic pollution. The rapid and high-precision harmonic detection and analysis technology will set the prerequisite for mitigating the harmonic crisis. This article presents a robust, stable, and real-time approach based on broad learning system (BLS) for detecting and analyzing the amplitudes of fundamental and harmonic components, which only requires half or even quarter cycle input. Additionally, a fast-converging adaptive parameter learning strategy using rectified adaptive moment estimation (RAdam) for BLS is proposed. To obtain the actual distorted signals, the signal acquisition system is built on the basis of the designed multi-pulse rectifier power supply (6/18 pulse). With the fundamental frequency deviation, interharmonics and noise, etc., the simulated signals synthesized by MATLAB tool and the actual signals acquired from the built system are attained to verify the performance of methods. The proposed approach differentiates from other existing methods not only due to its superior precision and robustness in the presence of multitudinous interference factors but also its real-time performance because of low average computing time.