Development and Application of a Multi-Input Collisional-Radiative Model

The collisional-radiative model (CRM) method establishes a correspondence between plasma parameters, such as electron density and electron temperature, and the intensity of spontaneous emission spectra by simulating collisional-radiative processes. Through lookup table construction, plasma parameters can be inversely derived from measured emission spectra to achieve spectral diagnostics. Previously, due to the exponential scaling of required sampling points with the number of input parameters in CRM spectral diagnostics, researchers often minimized CRM inputs and adopted global calculation. Now advances in computational power and data-driven techniques enable the development of multi-input CRM for more accurate spectral diagnostic.

This study proposes a multi-input CRM based on the local field approximation and transport-sensitive (TS) /local chemistry (LC) particle classifications. This approach relies on the high collision-radiation frequency of LC particles, whose densities are governed by collisional-radiative equilibrium and negligible sensitivity to drift-diffusion transport. By coupling CRM with the electron Boltzmann equation, LC particle densities are determined using inputs of reduced electric field(E/N) and TS particle densities. This approach eliminates the reliance on approximate parameters, such as global diffusion coefficients and radiation escape factors, and enables accelerated simulations of plasmas containing multi-excited-state particles. Neural networks are employed to map high-dimensional CRM inputs to outputs, with the generalization capability significantly reducing the sampling density required for CRM spectral diagnostics.