To resolve the design conflict between low heat flux density requirements in superconducting motor multi-layer insulation material and structural compactness demands, this study proposes a multi-objective optimization method for multilayer insulation based on an improved layer by layer model (LBL) and non-dominated sorting genetic algorithm. First, we enhanced the conventional LBL model’s computational accuracy by incorporating key parameters—spacer optical properties, reflective screen aperture ratio, and a dynamic adaptive coefficient—derived from fundamental radiative, gaseous, and solid conductive heat transfer equations. Second, we constructed a variable-density MLI model with up to four distinct density zones, accounting for relative heat transfer contributions. Finally, employing a non-dominated sorting genetic algorithm with layer counts per density zone as design variables and the improved LBL model as fitness function, we optimized the system under layer-count constraints per zone and total layer count, yielding the Pareto frontier through population evolution. Based on this, we further analyzed the relationships governing MLI heat flux density in relation to three key design parameters:the number of density zones, layer count per density zone, and layer density. Concurrently, we assessed the regulatory effects of variable-density configurations on heat flux distribution. Results demonstrate that the optimized solutions span heat flux densities of 0.42 to 3.11 W/m2 and thicknesses of 5.5 to 43.0 mm, encompassing four configuration types:uniform-density layouts, and variable-density configurations with two, three, or four distinct density zones. By adjusting the number of density zones and the layer density and number of layers in the density zones, multi-layer insulation material optimization can be achieved, reducing the difficulty of subsequent construction.