Generative Artificial Intelligence Music Therapy for Emotion Regulation: A Multi-Level Representation and Dynamic Adaptation Framework

Systematic quantitative explanations of the multi-level evolution of emotion regulation remain limited in intelligent music-based intervention research. Traditional music interventions also have limited capacity for dynamic and precise emotion regulation. To address this gap, a generative artificial intelligence (GAI) music therapy framework is developed for multi-level emotion representation and dynamic adaptation. Quantitative transmission pathways are examined across physiological response, psychological perception, and cognitive processing. The framework represents the process from acoustic stimulation input to emotional homeostasis output and compares static fixed-music intervention with GAI-driven dynamic music intervention. The multi-level representation model achieves stable fitting performance, with absolute fitting errors across levels maintained within 0.002-0.005. Under the simulation conditions, the overall transmission efficiency of dynamic music intervention is higher than that of static fixed music intervention. At the 180 s intervention point, the difference in transmission efficiency reaches 0.14, while the difference at the physiological level remains within 0.11-0.14. After 500 training iterations, the steady-state loss decreases to 0.0009. The steady-state fluctuation range of the core parameters is 0.006, the span across parameter groups does not exceed 0.04, and the maximum deviation across repeated parameter fitting is 0.009. The absolute values of all pairwise parameter correlation coefficients remain below 0.12, indicating no apparent multicollinearity in the model solution. Compared with existing studies that mainly examine intervention effects and provide limited quantitative modeling of multi-level coupling, this study develops a serial physiological-psychological-cognitive framework for emotion representation and describes the sequential transmission process through which music stimulation is associated with emotional regulation. The simulation results indicate that dynamic intelligent music stimulation can improve multi-level representation states and may reduce negative emotional responses within the modeled system. The findings provide a quantitative modeling reference for the design and optimization of personalized intelligent emotion-regulation approaches in biomedical and mental health applications.