Intelligent design of colored passive cooling multilayer films using bidirectional neural networks and genetic algorithms

Abstract

Colored passive cooling combines vibrant coloration with passive cooling capabilities, attracting significant interest in sustainable energy applications. While nanostructured colored passive cooling designs show promise, achieving precise colors with cooling power remains computationally challenging due to complex geometric parameter optimization. This study presents an innovative bidirectional design framework combining bidirectional neural network (BNN) and genetic algorithm (GA), to assist in the design of multilayer films. BNN accurately forecasts color and cooling power (99.67 % accuracy) from structural parameters and temperature T, and inversely designs geometric parameters (99.86 % accuracy) based on desired color and cooling performance at the given temperature. Crucially, the GA-based framework explores multiple high-precision solutions based on desired parameters, effectively addressing the “one-to-many” inverse design problem, overcoming the BNN’s single-solution limitation. The designed PMMA/TiN/TiO₂/Ag films achieve a broad color gamut, covering 62 % of the CIE-1931 color space, while maintaining its equilibrium temperature only 2 −3 K above the ideal device. Together, these machine learning frameworks establish a full-cycle design paradigm: BNN enables bidirectional property-structure mapping with ultra-high accuracy while the GA- forward prediction model hybrid efficiently generates diverse optimal designs satisfying multi-objective constraints. This dual methodology accelerates the discovery of novel colored passive coolers, accelerating the development and deployment of energy-efficient solutions for significant contributions to energy conservation and sustainable development.