This code package consists of four parts: Genetic Algorithm (GA) code, training data, Convolutional Neural Network (CNN) code, and GA‑CNN code. Detailed descriptions are as follows:

(1) GA code: Through co‑simulation between MATLAB and CST, this code automates the modelling process and performs high‑quality simulation calculations for 13,022 data points. These simulation results are subsequently used for CNN training.

(2) Training data: Among the 13,022 data pairs, the input data consist of the encoding sequences for each coding metasurface (file: x_n36.txt), and the output data are the corresponding broadband reflection loss (RL) values obtained from simulation (file: y_n36.txt).

(3) CNN code: This code trains the CNN model using the aforementioned training data.

(4) GA‑CNN code: This code replaces the full‑wave simulation (CST) with the trained CNN model and performs large‑scale prediction of the reflection loss performance for coding metasurfaces.