Grid-connected hybrid solar PV–wind–battery system with AI-augmented energy management and robust inverter control for enhanced power quality: a simulation-based study

This paper presents an integrated modelling, control design and simulation-based validation of a three-phase grid-connected hybrid solar photovoltaic (PV)–wind energy conversion system supported by a lithium-ion battery energy storage system (BESS) and augmented with an artificial-intelligence (AI) energy-management layer. The central contribution is not the individual algorithms, which are established, but a co-designed, time-scale-separated architecture in which an offline hybrid Grey Wolf Optimizer–Differential Evolution (GWO–DE) tuner, a long short-term memory (LSTM) forecaster and a deep-deterministic-policy-gradient (DDPG) dispatcher are coupled to deterministic device-level loops through an explicitly bounded supervisory layer that preserves grid-code compliance even under AI failure. A 100 kW PV array is interfaced through a DC–DC boost converter governed by a GWO–DE-tuned variable-step Incremental Conductance maximum-power-point-tracking (MPPT) algorithm, while a 50 kW permanent-magnet synchronous generator wind subsystem feeds a common 800 V DC bus through an active rectifier with optimum-torque control. A 60 kWh BESS is integrated through a bidirectional converter and dispatched by a DDPG agent fed with LSTM forecasts of irradiance and wind speed; a coupled rain-flow degradation model quantifies the impact of the dispatch on battery life. The grid-side voltage source converter employs synchronous-reference-frame current control with a second-order generalized integrator phase-locked loop and an actively damped LCL filter. The complete plant has been modelled in MATLAB/Simulink and validated under both a 6 s sequential disturbance profile and a 24-h energy-management horizon. Results demonstrate 99.46% MPPT tracking efficiency, 28 ms settling time, 0.45% steady-state ripple, ±1.5% DC-link regulation, a full-bandwidth grid-current total harmonic distortion of 2.18% (1.60% up to the 50th order, the scope of IEEE-519) with 0.84% fifth-harmonic content, 32 ms low-voltage-ride-through recovery, a 14.8% reduction in daily grid energy import on the representative day (14.6% season-weighted), and a 9.5% reduction in levelized cost of energy compared with rule-based baselines, with the DDPG dispatcher additionally outperforming PPO, SAC and TD3 benchmarks under identical training budgets. The study is explicitly simulation-based; a detailed real-time computational mapping onto commercial-off-the-shelf embedded hardware is provided to establish implementation feasibility ahead of the planned hardware-in-the-loop validation.