Real-Time Parameters Estimations and Adaptive Control Structures for Power Converters in Grid-Tied and Railways Applications

This dissertation develops a unified framework for real time impedance awareness and adaptive control of power electronic converters operating under variable grid and load conditions. Variations in grid impedance, driven by changes in strength, topology, loading and operating points, directly influence stability margins, achievable bandwidth, harmonic behaviour and resonance phenomena, particularly in LCL filtered converters. To address these challenges, the thesis proposes an end to end workflow that connects measurement, impedance estimation, parameter extraction and automatic controller adaptation, with explicit consideration of disturbance magnitude, sensing resolution and computational latency. Three complementary estimation approaches are formulated. Active methods rely on controlled excitation: a band limited White Noise injection enables broadband impedance acquisition with minimal implementation overhead, while a deterministic Multi Sine signal provides frequency selective stimulation with improved signal to noise ratio and reduced observation time. Both methods are designed to operate within the dynamic limits of switching frequency, ensuring meaningful identification across the frequency range bounded by the LCL resonance. A Quasi Passive strategy is introduced to limit the frequency of active injections. Small variations of active and reactive power generate a compact set of operating points from which the impedance is estimated in real time. A triggering logic based on RMS voltage deviation and current harmonic distortion activates the estimator only when grid changes materially affect resistive or inductive behaviour. A fully passive method is also developed to exploit natural load variations, supported by a consistency metric that filters unreliable estimates in non ideal operating conditions. To support embedded implementation, each estimation method adopts processing steps aligned with its operating principle. The active estimators rely on frequency domain analysis to obtain broadband or selected frequency impedance information, enabling fast self commissioning of the current controller and configuration of the active damping layer. The Quasi Passive method instead uses sequential measurements derived from controlled active and reactive power variations and algebraic reconstruction in the dq frame, making it suitable for predictive control applications where low intrusiveness and selective triggering are required. A resonance tracking method is developed for railway systems. It operates on the DC side and identifies the dominant oscillatory mode directly from the power spectral density of the DC bus current, enabling selective resonances attenuation. The framework is validated through Hardware in the Loop real-time verifications and laboratory experiments on grid tied inverters, demonstrating consistent parameter identification across operating conditions and recovery of the intended control bandwidth after parameters updates. The results confirm improved stability in resonance affected scenarios and preservation of power quality performance. The railway application further shows that the same identification philosophy can be used to track time varying resonances in DC traction networks and to enhance stability under constant power load. Overall, the dissertation demonstrates that online grid parameters estimation, implemented through active, Quasi Passive, passive or resonance tracking approaches, becomes practically valuable when its outputs directly drive adaptive control actions, enabling stable and predictable operation in both AC grid tied converters and railway energy recovery systems.