The aircraft is often difficult to be stably evaluated due to energy fluctuations in the final approach phase. The traditional single-parameter threshold monitoring method is difficult to capture the complex coupling relationship between dynamic energy and potential energy, and the adaptability is insufficient under variable meteorological disturbances. Therefore, this study proposes a new multi-dimensional prediction and evaluation method, which integrates energy management theory and deep learning technology, aiming to improve the early recognition ability of unstable approach under complex meteorological conditions and optimize the energy regulation ability. Firstly, a new stability evaluation framework is constructed from the perspective of energy. Two core evaluation parameters of ' energy altitude ' and ' balance energy ' are proposed. This method breaks the traditional way of monitoring speed and altitude parameters in isolation. In this paper, a dynamic safety boundary function is designed based on the principle of flight mechanics and civil aviation specifications. The function uses an altitude attenuation mechanism to make the boundary shrink smoothly with the decrease of flight altitude. At the same time, the sliding window statistics and balanced energy triggering mechanism are introduced, which significantly enhances the adaptability of the boundary to various disturbances and effectively overcomes the lag problem of static boundary response. By establishing a multi-dimensional parameter system with energy altitude and balance energy as the core, this study reveals the mechanism of dynamic energy potential energy coupling on approach stability. The hybrid dynamic boundary function realizes the collaborative optimization of physical constraints and data-driven. The research results provide a new theoretical paradigm for solving the evaluation of unstable approach under complex weather, and have important theoretical value and engineering application prospects for ensuring flight safety.