During the handover process of high-speed trains, the reference signal receiving power (RSRP) experiences fluctuations caused by path loss and terrain, leading to issues with the traditional A3 event-based handover decision method, such as ping-pong occurrences and a decrease in handover success rate. To address this problem, a handover algorithm based on an improved gray-Markov model is proposed. This algorithm processes and predicts the received power of the reference signal using an enhanced grey-Markov model. The handover process then utilizes the prediction results as the basis for the handover decision based on the preloading method. Simulation results demonstrate that the improved handover algorithm significantly reduces the fluctuation of RSRP values received by the train, lowers the probability of ping-pong handovers, and effectively improves the handover success rate.