Deep reinforcement learning-based thermal-visual collaborative optimization control system for multi-sensory art installations

This paper presents an attention-enhanced deep reinforcement learning control system for thermal-visual collaborative optimization in multi-sensory art installations. The proposed system integrates three key innovations: (1) an attention-based DDPG algorithm with dynamic modality weighting (α = 0.6-0.8 for thermal, 0.6-0.7 for visual features), (2) adaptive sensor fusion combining Kalman and particle filtering with 8 ± 2 ms processing latency, and (3) hierarchical four-layer architecture achieving 65% control accuracy improvement and 40% response time reduction compared to traditional approaches. The system architecture incorporates a hierarchical four-layer design with perception, fusion, decision-making, and execution components. The core innovation lies in an attention-based deep reinforcement learning algorithm that dynamically processes multi-modal sensory inputs and optimizes thermal-visual coordination through continuous learning. The algorithm employs spatiotemporal alignment mechanisms, adaptive feature weighting, and collaborative optimization strategies to achieve superior control performance. Experimental validation demonstrates quantified improvements over conventional methods: control accuracy improved from 0.247 ± 0.067 RMSE (PID baseline) to 0.085 ± 0.012 RMSE (proposed method), representing 65% improvement; response times reduced from 78 ± 22 ms (fuzzy control) to 45 ± 8 ms (proposed method), achieving 40% improvement; energy efficiency increased from baseline consumption to 23% reduction through intelligent coordination. Real-world deployment results confirm practical effectiveness with measured user satisfaction scores of 4.1 ± 0.7 on a 5-point scale and system availability of 98.5% over 186-hour continuous operation periods. The proposed system enables more sophisticated multi-sensory experiences while maintaining artistic integrity and provides a foundation for advanced digital art technologies.