We study a class of regularization methods for solving least-squares ill-posed problem with a convex constraint. Convergence and convergence rate results are proven for the problems which satisfy so called power source condition. All the results are obtained under the assumptions that, instead of exact initial data, only their approximations are known.