<p>For solving the linear least squares problem, a maximal distance coordinate descent method is proposed, which adopts a greedy strategy on the basis of the distance criterion, and the theory of convergence for such a method in cases where the coefficient matrix is of full column rank has been shown. In addition, we introduce a new probability criterion, which is constructed based on the angle between hyperplanes such that the hyperplane with a larger angle to the current hyperplane can be selected for projection during the iteration process, construct a multi-step randomized coordinate descent method with new probability criterion and study its convergence. Finally, numerical experiments validate the effectivity of new methods.</p>