Application of ensemble machine learning algorithms and filtering techniques in slow orbit feedback systems of electron storage rings

Orbit stability is a critical performance metric for modern synchrotron radiation facilities and colliders, necessitating effective orbit correction and slow orbit feedback systems. In recent years, innovative orbit feedback methods leveraging neural networks have been progressively implemented, demonstrating improved performance. BEPCII has also attempted to implement an orbit feedback system based on neural networks. Furthermore, we have noted that orbit stability is significantly affected by noises. To address the impact of noises, we introduced the Kalman filtering method into the orbit feedback system and combined it with the existing neural network framework, creating a dual neural network orbit feedback system. This innovative approach has led to further improvements in overall orbit stability.