Data-driven evaluation and optimization of Touschek lifetime

Abstract The Touschek lifetime is one of the most important performance parameters of particle accelerators. The traditional methods for calculating the Touschek lifetime often involve substantial computational costs, especially during the design phase of the storage ring, where numerous alternative lattices need to be evaluated. In this paper, we’re proposing a fast machine-learning-based method to estimate the Touschek lifetime. This method tracks about 10% of the initial particles over the long term, trains a machine learning model to predict the survival or loss status of all the initial particles, and subsequently computes the Local Momentum Acceptance (LMA) to determine the Touschek lifetime. And the results based on High Energy Photon Source (HEPS) show that the method gets nearly 95% prediction accuracy for the particle survival status. It also reduces computation time by over 80% and shows strong adaptability across various lattices.