Wind turbine reliability is critically dependent on the health of rolling bearings. Traditional fault diagnosis methods often struggle to handle non-stationary and complex vibration signals. This paper proposes a hybrid intelligent diagnostic approach combining Adaptive Variational Mode Decomposition (AVMD), Improved Beluga Whale Optimization (IBWO), and Broad Learning System (BLS) to address this challenge under constant operating conditions. AVMD adaptively decomposes the vibration signals, while IBWO determines the optimal decomposition parameters and selects relevant IMFs based on energy difference. Multi-domain features are extracted and reduced using Kernel PCA before being fed into the IBWO-optimized BLS classifier. Experiments on publicly available datasets demonstrate that the proposed method significantly improves fault detection accuracy, interpretability, and computational efficiency.