In this study, an improved Variational Mode Decomposition method is introduced and applied to the multi-frequency estimation problem in heterophonical Turkish maqam music recordings. The main purpose of this method is to decompose a given signal into a predetermined, finite number of modes. Center frequencies of the fundamental components are calculated in a non-iterative and adaptive manner, while the modes are estimated simultaneously. In the traditional DKA algorithm, Tikhonov Regularization is employed to estimate the center frequencies of the modes. Instead, we propose to use ElasticNet Regression to solve the optimization problem and thus improve the fundamental frequency estimation performance. By using the proposed approach, simulations are performed on multiple frequency estimation in heterophonical Turkish maqam music recordings. Results are compared to that of the commonly used fundamental frequency estimation techniques in the literature such as Melodia and YIN, and evaluated by using MIREX2016-Multiple Fundamental Frequency Estimation and Tracking criteria. Experimental studies show that the proposed method improves the results of traditional VMD, and outperforms other methods used for comparison.