The acquisition of aeroacoustics signals of wind turbines is of great significance in environmental noise assessment and fault monitoring of blades. The single acoustic sensor is simpler and more flexible than the acoustic sensors array but it lacks spatial analysis capability of the acoustic pressure field, and it is difficult to get pure aeroacoustics signals directly. This paper proposes a single channel blind source separation (SCBSS) method based on variational mode decomposition (VMD) which is applied to the separation of the wind turbine aeroacoustics signals acquired by the single acoustic sensors. The variational mode decomposition of the nonlinear and nonstationary signals based on the data itself completely is adaptive. In addition, the problem of mode mixing and “endpoint effect” has been improved. A novel approach combined correlation criterion with an overall index of orthogonality criterion is proposed in this paper to determine the optimal number of decomposition layers of VMD. We transform single channel underdetermined blind source separation to the non-underdetermined problem by establishing virtual multi-channel signals of the observation signals base on VMD, and separate the signals by joint approximate diagonalization of eigen-matrices (JADE) of fourth-order cumulant matrices. The method proposed in this paper has an excellent separation performance for wind turbine aeroacoustics signals, and the analysis of simulation signals indicates it has a 92.23% average recognition proportion, which is better than BSS based on EMD and EEMD, and the method has an extremely shorter computing time than EEMDBSS. The analysis of actual signals shows that the suggested method is adaptive and robust for noise.