HOPFIELD NEURAL NETWORK METHOD BASED SHUNT ACTIVE POWER FILTER IN PV INTEGRATED GRID CONNECTED EV CHARGING SYSTEM

Abstract

This work compares two reference current estimation methods for a single-phase Shunt Active Power Filter (SAPF) applied to a photovoltaic (PV) integrated grid connected Electric Vehicle (EV) charging system. Diode-rectifier EV chargers intro duce harmonic currents at the point of common coupling (PCC); without compen sation, source current THD readily exceeds the 5% limit prescribed by IEEE Std 519-2014. The two methods examined are the Hopfield Neural Network (HNN) method and the Generalized Integrator (GI) method. In the HNN approach, load current is decomposed into in-phase and quadrature components whose estimation dynamics share a Lyapunov–Hopfield energy function, guaranteeing asymptotic stability. The GI uses a second-order resonant integrator tuned at 50 Hz to extract the fundamen tal component without phase lag; it is simpler to implement but its fixed resonant frequency reduces robustness under load transients. The full system, a 110 V, 50 Hz single-phase grid, diode-rectifier EV load, PV array with MPPT-controlled boost converter, 154 V lead-acid battery with bidirectional buck–boost converter, and a VSC-based SAPF was modelled in MATLAB/Simulink. Both controllers were tested under steady-state operation, a step reduction in irra diance from 1000 W/m2 to 500 W/m2 at t = 3.5 s, and a load step. The HNN achieves a source current THD of 3.61% against 4.35% for the GI, a 17.01% relative reduction, with both values within the IEEE 519-2014 limit. The HNN also shows tighter DC-link voltage regulation and faster settling under transients. The power balance holds throughout the 5 s simulation for both controllers. These results in dicate that the HNN is preferable where precise harmonic compensation is required; the GI remains a viable alternative under less demanding operating conditions.