The 0-1 multidimensional knapsack problem(MKP) is a classical NP-hard combinatorial optimization problem. In this paper, we propose a novel heuristic algorithm simulating evolutionary computation and large neighbourhood search for the MKP. It maintains a set of solutions and abstracts information from the solution set to generate good partial assignments. To find high-quality solutions, integer programming is employed to explore the promising search space specified by the good partial assignments. Extensive experimentation with commonly used benchmark sets shows that our approach outperforms the state of the art heuristic algorithms, TPTEA and DQPSO, in solution quality. It finds new lower bound for 8 large and hard instances