Induced Voltage Noise Compensation Method Based on CWW + CNN-BiLSTM-Attention

Abstract

Induced voltage noise is a significant source of interference in voltage detection, seriously affecting measurement accuracy and system stability. In the complex electromagnetic environment of the Tokamak system, induced voltage noise can exceed 500 V, posing a great challenge to noise suppression in voltage detection signals. This article presents a two-stage noise compensation approach designed to address the problem. The first stage uses a hardware-based co-wound wire (CWW) primary compensation method, effectively reducing the induced voltage noise from over 500 V to below 10 V. Building on this foundation, the second stage introduces a deep learning model based on CNN-BiLSTM-Attention as the secondary compensation module, which further reduces the residual noise to below 100 mV over a 1-s window. The proposed method was validated through real-time deployment on the Experimental Advanced Superconducting Tokamak (EAST). Experimental results demonstrate that the method not only achieves excellent compensation effect but also preserves the integrity of critical voltage detection signals throughout operation.