To address the engineering challenge of severely degraded defect localization accuracy in directly buried cables due to intense noise interference in the complex noise environment of wind and photovoltaic power stations, a frequency domain reflection (FDR) defect precise positioning method based on multi-physical-field noise coupling mechanism modeling and a two-stage joint noise reduction approach of ensemble empirical mode decomposition (EEMD)-singular value decomposition (SVD) was proposed. First, a distributed parameter multi-physics coupling model of cables integrating mechanical vibration and electromagnetic pulse interference was established, theoretically revealing the coupling pathways and influence mechanisms of composite noise on traveling wave signals. On this basis, a two-stage joint denoising algorithm of EEMD and SVD was proposed, the EEMD stage adaptively separated broadband noise in cable signals, while the SVD stage reconstructed the dominant eigenmode function with rank constraints, further effectively suppressing residual narrowband noise in cable signals and achieving high fidelity signal recovery. Simulation and experimental results show that this method can still improve the accuracy of cable defect localization to over 99.7% in strong noise backgrounds with signal-to-noise ratios as low as 0 dB, with an average localization error of less than 1 m. It also exhibits good adaptability and robustness to different degrees of defects. The research results provide an effective technical approach for precise positioning of cable defects in strong noise environments such as new energy stations. The proposed noise modeling and joint noise reduction framework has clear engineering promotion value.