Modeling and simulation optimization of interactive design systems based on artistic style transfer

This paper tackles key challenges in interactive design systems: high latency, weak user control, and the aesthetic-function trade-off. We propose an optimization method integrating lightweight generative networks with dynamic modeling. First, a feedforward network architecture based on the MobileNetV3 encoder and the AdaIN (Adaptive Instance Normalization) decoder is designed to achieve millisecond-level style transfer. Second, based on probabilistic state-space modeling theory, a human-machine collaborative state machine is constructed. This Markov decision process describes the transition probabilities of user operation sequences and integrates hard constraints such as readability and layout rationality. Then, a user-system co-simulation framework is proposed. A virtual user behavior simulator generates diverse interaction sequences, driving the NSGA-III (Non-dominated Sorting Genetic Algorithm III) algorithm to perform multi-objective optimization on style quality, response latency, and constraint satisfaction. Experimental results demonstrate significant improvements over baseline methods (AdaIN, WCT2) in the system's consistency of style expression, real-time interaction, and design usability. After 50 generations of optimization, the average FID (Fréchet Inception Distance) value drops from 20.4 to 13.5; the interaction latency decreases from 286 ms to 187 ms; and the constraint violation rate drops from 22.3% to 5.7%, a decrease of 16.6 percentage points, validating the effectiveness of the “modeling-simulation-optimization” methodology. This method achieves a closed-loop collaboration between art generation and engineering design, providing a modeling, simulation, and optimization solution for intelligent interactive design systems.