Abstract: Objectives: The massive deployment of the Starlink constellation occupies a substantial share of low Earth orbit (LEO) resources, increasing the risks of space traffic congestion and collisions. Analysis of its orbital maneuver strategies can provide data support for space traffic management and space situational awareness. Methods: An energy-based maneuver detection method derived from the conservation of specific mechanical energy is proposed utilizing publicly released Starlink precise ephemerides. To accurately calculate the orbital energy, the 120-order SGG-UGM-2 gravity field model is applied as the background dynamic model. By differencing the energy time series between adjacent epochs, maneuvers are identified using optimized parameters, specifically an energy variation threshold of 15 m²/s² and a minimum duration threshold of 2 minutes. The method can preserve the original one-minute temporal resolution of the ephemerides without data filtering or smoothing. Utilizing Two-Line Element (TLE) data and public databases, the Starlink constellation is categorized into distinct orbital shells based on orbital altitude and inclination, detailing the deployment status and satellite versions for each Generation 1 (Gen1) shell. Based on precise ephemerides data of 344 Starlink satellites selected from these shells from April to May 2024, orbital maneuvers during the operational, orbit-raising, parking, and descent phases are systematically identified and quantitatively analyzed. Results: 1) During the operational phase, Gen1 shell 1 satellites primarily maneuver once every two days, exhibiting a mean acceleration of 2.1×10^-4 m/s² with durations ranging from 2 to 25 minutes. Satellites in the other four Gen1 shells generally maneuver once per day, showing a mean acceleration of 2.2×10^-4 m/s² and durations between 2 and 15 minutes. During the maneuvering phase, the midpoints of the sub-satellite ground tracks are primarily distributed near the equator. 2) During the orbit-raising phase, compared to the maneuver frequency of approximately 0.1 days per maneuver reported in previous studies, the proposed energy-based method reveals a maneuver frequency of approximately 0.04 days per maneuver for most satellites, while only a small subset of early-version satellites exhibit a frequency of 0.1 days per maneuver. 3) During the parking phase, maneuvers occur every 0.2 to 0.3 days. 4) During the descent phase, maneuvers are less frequent and not fixed. The mean acceleration of the orbitraising, parking and descent phase is all 2.1×10^-4 m/s². Conclusions: The energy-based method preserves the original temporal resolution of the precise ephemerides, effectively detecting maneuvers, especially high-frequency, short-duration ones. The implementation of the energy-based method reveals distinct maneuver strategies of the Starlink constellation across different phases and shells.These strategies and the energy-based method can offer critical reference parameters for enhancing orbital safety, collision avoidance predictions, and global space traffic management in densely populated LEO environments.