For the 0-1 multi-dimensional knapsack problem (MdKP), a hybrid distribution estimation algorithm using the Markov network (hDEUM) is proposed, which uses a Markov random field as the probability distribution model and represents the relationships between variables with an undirected graph. To effectively rectify the infeasible solutions in the post-sampling population, a repair mechanism and a local enhancement operator are designed. Additionally, a neighborhood search operator is proposed to enhance the algorithm's local search capability. Experimental results on several standard benchmark sets demonstrate that the proposed hDEUM is effective and outperforms several existing evolutionary algorithms designed for solving the MdKP, validating its superiority for solving the 0-1 MdKP.