A Predefined-Time Zeroing Neural Network Model for Harmonic Detection in Modern Power Systems

Abstract

To satisfy the stringent real-time requirements of harmonic detection, this paper proposes a predefined-time zeroing neural network (PTZNN) model, marking the first application of the zeroing neural network (ZNN) model in modern power systems analysis. Compared with other detection methods, the PTZNN model integrates a novel predefined-time activation function (NPTAF), ensuring that the detection error converges to zero within a predefined time, thereby achieving predefined-time convergence. In addition, the PTZNN model eliminates the reliance on analytical derivatives via the proposed NPTAF, enabling high-accuracy harmonic detection even in the presence of numerical differentiation deviations. Comparative simulations confirm the effectiveness and superiority of the PTZNN model for harmonic detection under both steady-state and transient distorted signals, which cover typical operating scenarios in modern power systems. Finally, a real-time experimental platform based on RT-LAB and OP5600 is established to validate the effectiveness and engineering feasibility of the proposed PTZNN model for harmonic detection. Note to Practitioners—This study is motivated by the urgent need for rapid and precise harmonic detection in modern power systems. Conventional harmonic detection methods often face challenges in balancing processing speed with tracking accuracy, especially when signals are subjected to abrupt disturbances. The PTZNN model is presented to address these challenges, offering a significant practical advantage where practitioners can set a specific predefined time to ensure the harmonic detection error converges to zero by this deadline. This capability is critical for time-sensitive power quality compensation and protection. Additionally, the model remains effective even when facing numerical differentiation deviations, ensuring reliable performance on standard industrial processors that may lack high-precision computing power. The proposed framework is suitable for integration into power quality analyzers and the control loops of active power filters to provide timely data for power systems monitoring and compensation. It is particularly effective for tracking both steady-state and transient harmonics. However, it should be noted that the predefined convergence time must be balanced against the hardware sampling frequency. Setting an unrealistically short time may increase the computational load on the processor.