Chiral metasurfaces based on bound states in the continuum provide an important way to develop high-performance chiral optical devices. However, traditional design methods usually rely on complicated parameter scanning, which are limited by insufficient flexibility and easy trapping into local optima, and thus hardly achieve the optimal chiral optical response.This study adopts the double scythe-shaped (DSS) structure and proposes a hybrid inverse design strategy combining neural network and differential evolution algorithm (DE). It can efficiently and accurately design chiral metasurfaces with designated circular dichroism (CD) responses in the wavelength range of 1200—1500 nm.This model consists of a peak predictor and a spectrum predictor, which realizes high-precision and cross-dimensional mapping from structural parameters to chiral optical responses. It can accurately capture the key characteristics of CD spectra and exhibits excellent spectral prediction capability. On this basis, leveraging the global search ability of the DE algorithm, the optimal combination of structural parameters is inversely predicted driven by the target CD value.Experimental results show that the coefficient of determination (R²) of the two-stage neural network for structural parameter prediction exceeds 0.9899, and the R² values for wavelength and CD prediction are 0.8721 and 0.7631, respectively. The model can accurately restore the core characteristics of circular dichroism spectra and effectively address the strong nonlinear mapping problem between structural parameters and optical responses. The proposed model only requires 350 groups of simulation data for training, which is far less than the thousands of datasets required by existing deep learning methods.Compared with the existing inverse design methods, the proposed strategy achieves a higher CD value (≈0.9) and exhibits obvious advantages in both efficiency and performance.This study gives full play to the fast surrogate prediction advantage of neural networks and the global optimization capability of differential evolution algorithm. It breaks through the technical bottlenecks of traditional chiral metasurface design, opens up a new technical route for high-sensitivity chiral sensing applications, and provides a reference for the inverse design of other nanophotonic devices.