The frequency-based method relying on the Effective Vibration Length (EVL) can effectively circumvent the complex boundary effects of short cables and accurately identify their tension forces. Existing real-time cable tension identification methods based on EVL assume that EVL remains constant as cable tension varies, and thus the real-time tension can be determined by identified frequencies. However, the assumption of constant EVL lacks sufficient discussion and analysis. To address this, this paper constructs a finite element numerical model of a short cable with complex boundary conditions to explore the variation patterns of EVL under different tension states. It is found that significant changes in cable tension may lead to notable fluctuations in EVL, thereby affecting the accuracy of real-time tension identification. To improve the accuracy of real-time tension identification for short cables, this paper combines tension-EVL values from several working conditions and corrects EVL through spline function interpolation. When the tension varies by ±60% of the average tension (1.356 kN to 4.096 kN), theoretical real-time frequencies are used to identify real-time tensions. The absolute error range of identified tensions without EVL correction is -0.025×103 kN to -0.245×103 kN, while the absolute error range after EVL correction is -0.06×103 kN to -0.13×103 kN, demonstrating improved theoretical identification accuracy and stability. To extract real-time frequencies, this paper compares four time-frequency analysis methods and finds that Synchrosqueezing Transform (SST) offers the best frequency resolution and accuracy. Short-Time Fourier Transform, Continuous Wavelet Transform, and Hilbert-Huang Transform exhibit low time-frequency resolution and poor real-time frequency identification accuracy. Based on SST-acquired real-time frequencies and with EVL correction, the absolute error of the obtained real-time tensions is reduced by 50% compared to that without correction, validating the effectiveness and accuracy of the proposed method.