For global optimization problems with high dimension, the multi-scale quantum harmonic oscillator algorithm is hard to converge. For this problem, a covariance-matrix multi-scale quantum harmonic oscillator algorithm is proposed, and the mathematical model of core part is given, which improves the method of generating covariance matrix from the estimation of multivariate normal algorithm and reserves the memory of old sampling points. Moreover, dynamic iteration steps are intraduced to accelerate updating of the new covariance matrix. The experimental results show that the performance of the proposed algorithm is far better than that of the original algorithm, and it's obviously superior to four classic optimization algorithms on convergence precision, convergence rate and robustness.