This paper presents an innovative approach to constructing a digital twin for energy storage converter control using a constrained neural network model. The proposed method incorporates physical model stability constraints, significantly reducing the data required for accurate digital twin development. Traditional approaches often demand extensive datasets, which can be impractical and computationally intensive. By leveraging the stability characteristics of the physical model, our method ensures precise and reliable digital twin construction, enabling effective identification and optimization of control parameters. Metropolis-Hastings sampling is adopted to process the constraints. Simulation results demonstrate that our approach not only maintains high accuracy but also enhances efficiency and reliability, offering substantial improvements for practical applications in energy storage converter control.